TWI748201B - Entropy coding of transform coefficients suitable for dependent scalar quantization - Google Patents

Abstract

Concepts for transform coefficient block coding are described which enable coding of coefficients of a transform block in a manner suitable for dependent quantization and effectively implementable by entropy coding in terms of coding efficiency.

Description

Entropy coding technology for transform coefficients suitable for dependent scalar quantization

Invention field

This application is related to entropy coding such as the transformation coefficient level used for coding images or videos.

Background of the invention

When setting the quantization parameters, the encoder must consider comprehensively. Coarse rendering quantization will reduce the bit rate but increase the quantization distortion, while finer rendering quantization will reduce the distortion but increase the bit rate. It will be advantageous to arrive at the concept of improving coding efficiency for a given domain of available quantization levels. One such possibility is to use dependent quantization, where the quantization depends on the previously quantized and coded data and is continuously adjusted, but the dependence of the quantization also affects the interrelationship between the data items to be quantified and coded, and therefore affects Information on the availability of context-adaptive entropy coding. It would be advantageous to obtain a concept that realizes the coding of the coefficients of the transform block in a way that is suitable for dependent quantization and can be effectively implemented by entropy coding in terms of coding efficiency.

Summary of the invention

The object of the present invention is to provide this concept for coding transform coefficient blocks.

This goal is achieved by the subject matter of the independent claim of this application.

10: Transform block

12: transform coefficient

14: sub-block

16, 28, 30: Decoding/task

18, 24', 24": Decoding

20: The first scan pass/effective value/coordinate pass

20', 20": times

22: The second time / the level of the flag again and again

22': first scan pass

22": second scan pass

26: Third pass/scan pass

27: The third time

32, 34: check

40: update

42: Assumption

44: Arrow/code order

45: Find

46: State

50: predetermined transformation coefficient

51: Adjacent transform coefficient

52: Partial template

The favorable aspect is the subject of the subsidiary claim. Hereinafter, a preferred embodiment of the application will be described with reference to the figures. In the figure: Figure 1 shows an illustrative example of an image encoder which can be embodied as an operation according to any of the embodiments described below The block diagram of the video encoder.

Figure 2 shows (a) a transform encoder; and (b) a block diagram of a transform decoder to illustrate the basic method of block-based transform coding; Figure 3 shows a histogram illustrating the distribution of uniformly reconstructed quantizers.

Figure 4 shows (a) a schematic diagram of a transform block subdivided into sub-blocks and (b) sub-blocks to illustrate an example of scanning for transform coefficient levels, which is illustratively used in H.265|MPEG- An example of H HEVC; in particular, (a) show the 16×16 transform block divided into 4×4 sub-blocks and the coding order of the sub-blocks; (b) show the internal code of the 4×4 sub-blocks The coding order of the transform coefficient level. Subdivision can be exemplarily used in the embodiments of the present application, used for the pass of coefficients when decoding the flags and remainders of coefficients, and used for state transition when dequantizing these coefficients.

Figure 5 shows the multi-dimensional output space spanned by one axis for each transform coefficient and the allowable reconstruction vector for the simple conditions of the following two transform coefficients Schematic diagram of the position: (a) Non-dependent scalar quantification; (b) An example of dependent scalar quantification.

Figure 6 shows a block diagram of a transform decoder that uses dependent scalar quantization to form an embodiment of the media decoder according to the present application. The modification of conventional transform coding (using non-dependent scalar quantizer) can be derived by comparing with Fig. 2b. Correspondingly, an embodiment using dependent scalar quantization coding transformation block can be obtained by modifying the encoder of FIG. 2a as well.

FIG. 7 is a schematic diagram illustrating an embodiment of two dependent quantization sets of a reconstruction level determined entirely by a single quantization step size Δ. The two available sets of the reconstruction level are highlighted as set 0 (top line) and set 1 (bottom line). An example of a quantization index indicating the reconstruction level within the set is given by the number below the circle. Open circles and filled circles indicate two different subsets within the set of reconstruction levels; these subsets can be used to determine the set of reconstruction levels of the next transform coefficient in the reconstruction order. Both sets include reconstruction levels equal to zero, but are otherwise disjoint; the two sets are symmetrical around zero.

之值。變數setId[k]指定適用於當前變換係數之重構層級之集合。其係基於重構次序中之在前變換係數來判定；setId[k]之可能值為0及1。變數n指定量化步長之整數因子；其係由重構層級之所選集合(亦即，setId[k]之值)及所傳輸之量化索引level[k]給定。 Figure 8 shows a pseudo code illustrating an example of the reconstruction procedure of transform coefficients. k represents the index that specifies the reconstruction order of the current transform coefficient, the quantization index of the current transform coefficient is represented by level[k], the quantization step size △ _k applicable to the current transform coefficient is represented by quant_step_size[k], and trec[k ] Represents the reconstructed transform coefficient

The value. The variable setId[k] specifies the set of reconstruction levels applicable to the current transform coefficient. It is determined based on the previous transform coefficient in the reconstruction order; the possible values of setId[k] are 0 and 1. The variable n specifies an integer factor of the quantization step size; it is given by the selected set of reconstruction levels (ie, the value of setId[k]) and the transmitted quantization index level[k].

scale[k]．2^-shift 之方式來選擇。 Figure 9 shows a pseudo code illustrating an alternative implementation of the pseudo code in Figure 8. The main change is to use integer implementation of scale and shift parameters to represent multiplication with quantization step size. Generally, the shift parameter (represented by shift) is constant for the transform block, and only the scale parameter (given by scale[k]) can depend on the position of the transform coefficient. The variable add represents the rounding offset, which is usually set equal to add=(1<<(shift-1)). Wherein △ _k is the nominal transformation ratio of the quantization step, and the shift parameter scale [k] based △ _k to obtain I

scale [ k ]. 2 ^- The way of ^{shift to choose.}

Figure 10 shows a schematic diagram of an example of splitting the set of reconstruction levels into two subsets. The two quantized sets shown are the quantized sets of the example in FIG. 7. The two subsets of quantized set 0 are labeled with "A" and "B", and the two subsets of quantized set 1 are labeled with "C" and "D".

Fig. 11 shows a pseudo code illustrating an example of a reconstruction procedure of transform coefficients of a transform block. The array level represents the transmitted transform coefficient level (quantization index) of the transform block, and the array trec represents the corresponding reconstructed transform coefficient. The 2d table state_trans_table specifies the state transition table, and the table setId specifies the quantized set associated with the state.

FIG. 12 shows a schematic diagram illustrating the state transition of dependent scalar quantization as a trellis structure. The existing horizontal lines represent different transform coefficients in the reconstruction order. The vertical axis represents the different possible states in the dependent quantization and reconstruction procedures. The connections shown specify the available paths between the states of different transform coefficients.

Figure 13 shows an example of a basic grid cell.

FIG. 14 shows a schematic diagram of the transform block used to explain the position of the first non-zero quantization index transmitted in the coding order, and the position is obtained by backfilling (back filling) description. Except for the position of the first non-zero transform coefficient, only the bin of the shadow coefficient is transmitted, and it is inferred that the coefficient marked in white is equal to zero.

Figure 15 shows a pseudo code illustrating a comparative example of coding at the transform coefficient level of a coefficient block, such as a block that can be easily transferred to decode transform blocks that are not divided into sub-blocks and transfer to coefficient decoding (also That is, by replacing the sub-blocks of "code writing" with "decoding". Here, the on pass is used to write all the flags and remainders in the code coefficient except for the sign flags.

Figure 16 shows a schematic diagram of the transformation block and illustrates a template for the selection probability model. The black square designates the current scan position, and the shaded square represents the local neighborhood used to derive the context model.

Figure 17 shows a schematic diagram of an example grid structure that can be used to determine a sequence (or block) of quantization indexes that minimize cost metrics (such as the Lagrangian cost metric D+λ.R). Show a grid for 8 transform coefficients (or quantization indexes). The first state (on the extreme left) represents the initial state set equal to zero.

Figure 18 shows a pseudo code that illustrates an embodiment of coding the transform coefficient level of a coefficient block, such as a block that can be easily transferred to decode transform blocks that are not divided into sub-blocks and transfer to coefficient decoding (also That is, by replacing the sub-blocks of "code writing" with "decoding". Here, it is used to write all the flags and remainders in the code coefficient except for the symbol flags. Here, the common pass in Figure 15 is split into three passes.

Figure 19 shows the pseudo code for an embodiment of coding the transform coefficient level of the coefficient block, such as the block that can be easily transferred to decode the transform block that is not divided into sub-blocks and transfer to the decoding of the coefficients (also That is, by "Decode" replaces the sub-blocks of "Write code"). Here, it is used to write all the flags and remainders in the code coefficient except for the symbol flags. Here, the common pass in FIG. 15 is split into two passes, one pass is used for the remainder, and the other pass is used for flags other than the symbol flag.

Figure 20 shows the pseudo code for an embodiment of coding the transform coefficient level of the coefficient block, such as the block that can be easily transferred to decode the transform block that is not divided into sub-blocks and transfer to the decoding of the coefficients (also That is, by replacing the sub-blocks of "code writing" with "decoding". Here, it is used to write all the flags and remainders in the code coefficient except for the symbol flags. Here, the common pass in FIG. 15 is split into several passes, and the co-location flags are respectively coded before the absolute level or before any other flags.

Detailed description of the preferred embodiment

According to the embodiment described below, transform coefficient-level entropy coding is performed in a manner suitable for effective implementation and dependent quantization and context-adaptive entropy coding such as context-adaptive binary arithmetic coding. These embodiments are particularly advantageous for entropy coding of transform coefficient levels in the context of transform coding using dependent scalar quantization. However, these embodiments are also usable and advantageous when used in conjunction with the conventional non-dependent scalar quantification. That is, these embodiments are also suitable for entropy coding of transform coefficient levels in the context of transform coding using conventional non-dependent scalar quantization. In addition, the embodiments described below are suitable for supporting the switch between transform coding using dependent quantization and transform coding using conventional non-dependent quantization (for example, in sequence, image, tile, image block or region). Block level) encoding and decoding Device.

In the embodiment described below, the transform coding is used to transform the sample set. It can be implemented as dependent scalar quantization or alternatively implemented as non-dependent scalar quantization to quantize the resulting transform coefficients and perform entropy coding of the obtained quantization index. On the decoder side, the set of reconstructed samples is obtained by entropy decoding of quantization index, dependent reconstruction of transform coefficients (or alternatively, non-dependent reconstruction), and inverse transformation. The difference between dependent scalar quantization and conventional non-dependent scalar quantization is that for dependent scalar quantization, the set of allowable reconstruction levels of transform coefficients depends on the transmitted transform before the current transform coefficient in the reconstruction order Coefficient level. This aspect is used in entropy coding by using different sets of probability models with different sets of allowable reconstruction levels. In order to achieve high-efficiency hardware implementation, the binary decision (referred to as bit sub) related to the transformation coefficient level of the block or sub-block is written in multiple passes. The binarization of transform coefficient levels and the distribution of binary decision (also known as bit elements) over multiple passes are based on the data written in the first pass to uniquely determine the possibility of the next scan position. Allows the choice of the way to reconstruct the set of levels. This has the advantage that it may depend on the set of allowable reconstruction levels (for the corresponding transform coefficients) to select the probability model of a part of the bits in the first pass.

The description of the following embodiment mainly targets the lossy coding of blocks of prediction error samples in image and video codecs, but the embodiment can also be applied to other areas of lossy coding. In particular, there is no restriction on the set of samples forming a rectangular block, and no restriction on the set of samples representing the prediction error samples (that is, the difference between the original signal and the predicted signal).

All current advanced technology video codecs such as the international video coding standards H.264|MPEG-4 AVC[1] and H.265|MPEG-H HEVC[2] follow the basic method of hybrid video coding. The video image is divided into blocks. The samples of the blocks are predicted using intra-image prediction or inter-frame prediction, and the samples of the obtained prediction error signal (the difference between the original sample and the predicted signal sample) are coded using transform Write code.

Figure 1 shows a simplified block diagram of a typical modern video encoder. The video images of the video sequence are coded in a certain order, which is called the coding order. The order of image coding can be different from the order of capture and display. For actual coding, each video image is divided into blocks. A block contains samples of a rectangular area of a specific color component. The entity corresponding to the block of all color components in the same rectangular area is often called a unit. Depending on the purpose of block division, in H.265|MPEG-H HEVC, coding tree block (CTB), coding block (CB), prediction block (PB) and transform block can be distinguished (TB). The associated units are referred to as a coding tree unit (CTU), a coding unit (CU), a prediction unit (PU), and a transformation unit (TU).

Generally, a video image is initially divided into fixed-size units (ie, aligned fixed-size blocks for all color components). In H.265|MPEG-H HEVC, these fixed-size units are called Code Tree Units (CTU). Each CTU can be further split into multiple code writing units (CU). The code writing unit is the entity of the selected code writing mode (for example, code writing within or between images). In H.265|MPEG-H HEVC, the CTU is decomposed into one or more CUs which are specified by the Quadtree (QT) syntax and transmitted as part of the bit stream. The CU of the CTU is located in the so-called z-scan order reason. It means that the four blocks generated by splitting are processed in raster scan order; and if any of the blocks is further divided, the corresponding four blocks (including the smaller blocks included) are being processed Processing is performed before a block under a higher split level.

If the CU is coded in the in-frame coding mode, the intra-frame prediction mode for the luma signal is transmitted, and if the video signal includes chrominance components, another intra-frame prediction mode for the chrominance signal is transmitted. In ITU-T H.265|MPEG-H HEVC, if the CU size is equal to the minimum CU size (as signaled in the sequence parameter set), the brightness block can also be split into four blocks of equal size. In this case , For each of these blocks, a separate brightness intra-frame prediction mode is transmitted. The actual intra-frame prediction and code writing are based on the transformation block. For each transformation block of the CU coded in the image, the reconstructed samples of the same color component are used to derive the prediction signal. The algorithm used to generate the prediction signal of the transformed block is determined by the transmitted intra-frame prediction mode.

In the inter-picture coding mode, the coded CU can be further split into multiple prediction units (PU). The prediction unit is the entity of brightness, and for color video, two associated chroma blocks (covering the same image area) using a single prediction parameter set. The CU can be coded as a single prediction unit, or it can be split into two non-square prediction units (supporting symmetric and asymmetric splitting) or four square prediction units. For each PU, an individual motion parameter set is transmitted. Each motion parameter set includes several motion hypotheses (one or two motion hypotheses in H.265|MPEG-H HEVC), and for each motion hypothesis, it includes a reference image (via the reference image in the reference image list) Index indication) And the associated motion vector. In addition, H.265|MPEG-H HEVC provides a so-called merge mode, in which the motion parameters are not explicitly transmitted, but are derived based on the motion parameters of adjacent blocks in space or time. If the CU or PU is coded in the merge mode, it will only be transmitted to the index in the motion parameter candidate list (this list is derived using the motion data of space and time adjacent blocks). The index completely determines the set of motion parameters used. The prediction signal of the PU coded between frames is formed by the prediction of motion compensation. For each motion hypothesis (specified by the reference image and motion vector), the prediction signal is formed by the displacement block in the specified reference image, where the displacement relative to the current PU is specified by the motion vector. The displacement is usually specified by the accuracy of the sub-sample (in H.265|MPEG-H HEVC, the motion vector has the accuracy of a quarter of the brightness sample). For non-integer motion vectors, the prediction signal is generated by interpolating the reconstructed reference image (usually using a separable FIR filter). The final prediction signal of the PU using multiple hypothesis prediction is formed by the weighted sum of the prediction signals of the individual motion hypotheses. Generally, the same motion parameter set is used for the lightness and chroma blocks of the PU. Even if the current advanced video coding standards use translational displacement vectors to specify the motion of the current region (block of samples) relative to the reference image, it is possible to use higher-order motion models (for example, affine motion models). In that situation, additional motion parameters must be transmitted for motion assumptions.

For both CUs that are coded within and between images, the prediction error signal (also referred to as the residual signal) is usually transmitted via transform coding. In H.265|MPEG-H HEVC, the block of luma residual samples and the block of chroma residual samples (if any) of the CU are divided into transform blocks (TB). The division of CU into transform blocks is indicated by the quadtree syntax. The four The fork tree syntax is also called residual quadtree (RQT). The resultant transform block uses transform coding to write codes: apply 2d transform to the block of residual samples, use non-dependent scalar quantization to quantize the transform coefficients, and perform entropy on the transform coefficient level (quantization index) obtained Write code. In the P and B blocks, skip_flag is transmitted at the beginning of the CU syntax. If this flag is equal to 1, it indicates that the corresponding CU is composed of a single prediction unit coded in the merge mode (ie, inferred that merge_flag is equal to 1) and all transform coefficients are equal to zero (ie, the reconstructed signal is equal to the predicted signal) . In that case, except for skip_flag, only merge_idx is transmitted. If skip_flag is equal to 0, then the prediction mode (inter-frame or in-frame) is signaled, followed by the grammatical features described above.

Since the coded image can be used for the motion compensation prediction of the block in the following image, the image must be completely reconstructed in the encoder. Add the reconstructed prediction error signal of the block (obtained by reconstructing the transform coefficient of the given quantitative index and the inverse transform) to the corresponding prediction signal and write the result into the buffer of the current image. After all blocks of the image are reconstructed, one or more in-loop filters (for example, deblocking filters and sample adaptive offset filters) can be applied. The final reconstructed image is then stored in the decoded image buffer.

The embodiments described below present the concept of transform coding, such as transform coding of prediction error signals. This concept is applicable to both in-image and inter-image coding blocks. It is also suitable for the conversion and coding of non-rectangular sample areas. Compared with conventional transform coding, according to the embodiment described below, transform coefficients are not quantized non-dependently. It applies at least to the use of dependent quantities To quantify. According to dependent quantization, the set of available reconstruction levels for a particular transform coefficient depends on the selected quantization index of other transform coefficients. The following describes the modification of the entropy coding of the quantization index, which improves the coding efficiency and maintains the ability to combine with dependent scalar quantization.

All major video coding standards (including the current advanced technology standard H.265|MPEG-H HEVC) use the concept of transform coding to code blocks of prediction error samples. The prediction error sample of the block represents the difference between the sample of the original signal and the sample of the prediction signal of the block. The prediction signal is by intra-image prediction (in this case, the samples of the prediction signal of the current block are derived based on the reconstructed samples of adjacent blocks in the same image) or by inter-image prediction (in In this case, the samples of the prediction signal are derived based on the samples of the reconstructed image). The sample of the original prediction error signal is obtained by subtracting the sample value of the prediction signal from the sample value of the original signal of the current block.

The transformation code of the sample block can be composed of the linear transformation of the quantization index, the scalar quantization and the entropy code. On the encoder side (see Figure 2a), a linear analysis transform A is used to transform the N×M block of the original sample. The result is an N×M block of transform coefficients. The transform coefficient t _k represents the original prediction error samples in different signal spaces (or different coordinate systems). The N×M transform coefficients are quantized using an N×M non-dependent scalar quantizer. Each transform coefficient t _{k is} mapped to a quantization index q _k , and the quantization index is also called a transform coefficient level. The obtained quantization index q _k is entropy-coded and written into the bit stream.

On the decoder side depicted in Figure 2b, the transform coefficient level q _{k is} decoded from the received bit stream. Each transform coefficient levels q _k mapped to the reconstructed transform coefficients t _'k. The N×M blocks of reconstructed samples are obtained by transforming the blocks of reconstructed transform coefficients using linear composite transform B.

Even if the video coding standard only specifies synthetic transformation B , the convention is to use the inverse transformation of synthetic transformation B as the analytical transformation A in the encoder, that is, A = B ^-1 . In addition, the transformation used in the actual video coding system represents an orthogonal transformation ( B ^-1 = B ^T ) or a transformation close to orthogonal. For orthogonal transform, the mean square error (MSE) distortion in the signal space is equal to the MSE distortion in the transform domain. Orthogonality has the following important advantages: a non-dependent scalar quantizer can be used to minimize the MSE distortion between the original sample block and the reconstructed sample block. Even if the actual quantization procedure used in the encoder considers the dependence between the transformation coefficient levels (introduced by the entropy code description above), the use of orthogonal transformation also significantly simplifies the quantization algorithm.

For a typical prediction error signal, transformation has the following effect: the signal energy is concentrated in several transformation coefficients. Compared with the original prediction error samples, the statistical dependence between the obtained transform coefficients is reduced.

In the current advanced technology video coding standard, the separable discrete cosine transform (Type II) or its integer approximation is used. However, the transformation can be easily replaced without modifying other aspects of the transformation coding system. Examples of improvements that have been proposed in the literature or standardization documents include:

. The use of Discrete Sine Transform (DST) for intra-image prediction blocks (may depend on the intra-frame prediction mode and/or block size). It should be noted that H.265|MPEG-H HEVC already includes DST for 4×4 transform blocks for intra-picture prediction.

. Switching transformation: the encoder chooses among a set of predefined transformations actually Use the transformation. The set of available transforms is known to both the encoder and the decoder so that it can use the index into the list of available transforms to communicate efficiently. The set of available transforms and their order in the list may depend on other coding parameters of the block, such as the selected intra-frame prediction mode. Under special circumstances, the transformation used is completely determined by coding parameters such as the intra-frame prediction mode, so that there is no need to transmit syntax elements for specifying the transformation.

. Inseparable transform: The transform used in the encoder and decoder means inseparable transform. It should be noted that the concept of switching transformations may include one or more inseparable transformations. Due to complexity, the use of inseparable transforms may be limited to certain block sizes.

. Multi-level transformation: The actual transformation consists of two or more transformation stages. The first transformation stage can be composed of separable transformations with low computational complexity. And in the second stage, an inseparable transform is used to further transform a subset of the obtained transform coefficients. Compared with the inseparable transformation used for the entire transformation block, the two-stage method has the advantage of applying more complex inseparable transformation to a smaller number of samples. The concept of multi-level transformation can be efficiently combined with the concept of switching transformation.

A scalar quantizer is used to quantize the transform coefficients. As a result of quantization, the set of allowable values of transform coefficients is reduced. In other words, the transform coefficients are mapped to a countable set (actually, a finite set) of the so-called reconstruction level. The set of reconstruction levels represents an appropriate subset of the set of possible transform coefficient values. In order to simplify the following entropy coding, the permissible reconstruction levels are represented by quantization indexes (also called transform coefficient levels), which are transmitted as part of the bit stream. On the decoder side, quantization index (transform coefficient level) mapping To reconstructed transform coefficients. The possible values of the reconstructed transform coefficients correspond to the set of reconstruction levels. On the encoder side, the result of scalar quantization is a block of transform coefficient level (quantization index).

In the current advanced technology video coding standard, uniform reconstruction quantizer (URQ) is used. The basic design is illustrated in Figure 3. URQ has the following properties: the reconstruction levels are equally spaced. The distance Δ between two adjacent reconstruction levels is called the quantization step size. One of the reconstruction levels is equal to 0. Therefore, the complete set of available reconstruction levels is uniquely specified by the quantization step Δ. Quantization index q to the reconstructed transform coefficients t 'of the decoder mapping principles given by the following simple equation t' = q. △.

In this context, the term "independent scalar quantization" refers to the following property: given the quantization index q of any transform coefficient, the associated reconstructed transform can be determined independently of all the quantization indexes of other transform coefficients The coefficient t' .

Since video decoders usually use integer arithmetic with standard accuracy (for example, 32 bits), the actual formula used in the standard may be slightly different from simple multiplication. When ignoring the clipping of the supported dynamic range of the transform coefficients, the reconstructed transform coefficients in H.265|MPEG-H HEVC t' = (scale. q + (1≪( shift-1)))≫shift, where the operators "<<" and ">>" represent bit shifts to the left and right, respectively. When ignoring integer arithmetic, the quantization step size △ corresponds to the following term: △=scale. 2 ^-shift .

Early video coding standards such as H.262|MPEG-2 Video also specify a modified URQ so that the distance between reconstruction level zero and the first non-zero reconstruction level is increased relative to the nominal quantization step size (for example, Increase to three-half of the nominal quantization step size △).

The quantization step size (or scale and shift parameter) of the transform coefficient is determined by the following two factors: ‧Quantization parameter QP: The quantization step size can usually be modified based on the block. For that purpose, the video coding standard provides a predefined set of quantization steps. The quantization step used (or equivalently, the parameters "scale" and "shift" introduced above) are indicated by using the index in the predefined list of quantization steps. This index is called quantization parameter (QP). In H.265|MPEG-H HEVC, the relationship between QP and quantization step size is roughly given by the following formula

The tile QP is usually transmitted in the tile header. Generally speaking, it is possible to modify the quantization parameter QP based on the block. For that goal, the difference quantization parameter (DQP) can be transmitted. The quantization parameter used is determined by the transmitted DQP and the predicted QP value, which is derived using the QP of the coded (usually adjacent) block.

Quantization weighting matrix: Video coding standards usually provide the possibility of using different quantization step sizes for individual transform coefficients. This is achieved by specifying so-called quantization weighting matrices w , which can usually be selected by the encoder at the sequence or image level and transmitted as part of the bit stream. The quantization weighting matrix w has the same size as the corresponding block of the transform coefficient. The quantization step △ _{ik of the} transform coefficient t _ik is given by the following formula △ _ik = w _ik . △ _block , where △ _block represents the quantization step size of the block under consideration (indicated by the block quantization parameter QP), i and k represent the coordinates of the current transformation coefficient within the specified transformation block, and w _ik represents the quantization weighting matrix w The corresponding input item.

The main purpose of the quantization weighting matrix is to provide the possibility for introducing quantization noise in a perceptually meaningful way. By using an appropriate weighting matrix, the spatial contrast sensitivity of human vision can be used to achieve a better trade-off between bit rate and subjective reconstruction quality. Nevertheless, many encoders use so-called flat quantization matrices (which can be efficiently transmitted using high-level syntax elements). In this case, the same quantization step size Δ is used for all transform coefficients in the block. Then, the quantization step size is fully specified by the quantization parameter QP.

The blocks of the transform coefficient level (quantization index of the transform coefficient) are entropy coded (that is, they are transmitted as part of the bit stream in a lossless manner). Since the linear transformation can only reduce the linear dependence, the entropy coding of the transformation coefficient level is usually designed in such a way that the remaining non-linear dependence between the transformation coefficient levels in the block can be used for efficient coding. Well-known examples are run-level coding in MPEG-2 Video, run-level-last coding in H.263 and MPEG-4 Visual, H.264|MPEG-4 Context-adaptive variable length coding (CAVLC) in AVC, and context-based adaptive binary arithmetic coding (CABAC) in H.264|MPEG-4 AVC and H.265|MPEG-H HEVC .

Current advanced technology video coding standards The CABAC specified in H.265|MPEG-H HEVC follows a general concept that can be applied to various transform block sizes. Transformation blocks larger than 4×4 samples are divided into 4×4 sub-blocks. Fig. 4a and Fig. 4b illustrate the division of an example of a 16×16 transform block. The coding order of the 4×4 sub-blocks shown in Fig. 4a and the coding order of the transform coefficient levels inside the sub-blocks shown in Fig. 4b are roughly based on the reverse diagonal scanning shown in the diagram Specify. For some intra-image prediction blocks, horizontal or vertical scanning mode (depending on the actual intra-frame prediction mode) is used. The write code order always starts at the high frequency position.

In H.265|MPEG-H HEVC, the transform coefficient hierarchy is transmitted based on 4×4 sub-blocks. Lossless coding of transform coefficient levels includes the following steps:

1. The transmission syntax element coded_block_flag, which signals whether there is any non-zero transform coefficient level in the transform block. If coded_block_flag is equal to 0, no further coding is performed on the data for the transformed block.

2. Transmit the x and y coordinates of the first non-zero transform coefficient level in the coding order (for example, the block-by-block reverse diagonal scanning order illustrated in FIG. 4). Split the transmission of coordinates into the first code and the last code part. The standard uses the syntax elements last_sig_coeff_x_prefix, last_sig_coeff_y_prefix, last_sig_coeff_x_suffix, and last_sig_coeff_x_suffix.

3. Start with the 4×4 sub-block containing the first non-zero transform coefficient level in the coding order, and process the 4×4 sub-blocks in the coding order. The coding of the sub-blocks includes the following main steps:

a. Transmission syntax element coded_sub_block_flag, which indicates the sub-area Whether the block contains any non-zero transform coefficient levels. For the first and last 4×4 sub-blocks (ie, sub-blocks containing the first non-zero transform coefficient level or DC level), this flag is not transmitted but it is inferred to be equal to one.

b. For all transform coefficient levels in the sub-block whose coded_sub_block_flag is equal to one, the syntax element significant_coeff_flag indicates whether the corresponding transform coefficient level is not equal to zero. If the value of this flag cannot be inferred based on the transmitted data, only this flag will be transmitted. In particular, if the DC coefficient is located in a sub-block different from the first non-zero coefficient (in the coding order) and all other significant_coeff_flags of the last sub-block are equal to zero, then the first important scan position (by transmitted x And the y coordinate designation) transmits the flag and does not transmit the flag for the DC coefficient.

c. For the eight transform coefficient levels before the significant_coeff_flag is equal to one (if any), the flag coeff_abs_level_greater1_flag is transmitted. It indicates whether the absolute value of the transform coefficient level is greater than one.

d. For the first transform coefficient level whose coeff_abs_level_greater1_flag is equal to one (if any), the flag coeff_abs_level_greater2_flag is transmitted. It indicates whether the absolute value of the transform coefficient level is greater than two.

e. For all levels where significant_coeff_flag is equal to one (exceptions are described below), the syntax element coeff_sign_flag is transmitted, which specifies the sign of the transform coefficient level.

f. The absolute value has not been completely determined by the values of significant_coeff_flag, coeff_abs_level_greater1_flag and coeff_abs_level_greater2_flag All transform coefficient levels that are specified (if any of the transmitted flags are equal to zero, the absolute value is completely specified), the remainder of the absolute value is transmitted using the multi-level syntax element coeff_abs_level_remaining.

In H.265|MPEG-H HEVC, all syntax elements are written using context-based adaptive binary arithmetic coding (CABAC). All non-binary syntax elements are first mapped to a series of binary decisions, which are also called bits. The obtained bit subsequence is written using binary arithmetic coding. For this goal, each element is associated with a probability model (binary probability mass function), which is also called a context. For most bits, the context represents an adaptive probability model, which means that the associated binary probability quality function is updated based on the bit value actually written. The conditional probability can be used by switching up and down for certain bits based on the transmitted data. CABAC also includes the so-called bypass mode, in which a fixed probability mass function (0.5, 0.5) is used.

The context of code selection for coded_sub_block_flag depends on the value of coded_sub_block_flag of adjacent sub-blocks that have been coded. The context of significant_coeff_flag is selected based on the scan position (x and y coordinates) within the sub-block, the size of the transform block, and the value of coded_sub_block_flag in the adjacent sub-block. For the flags coeff_abs_level_greater1_flag and coeff_abs_level_greater2_flag, the context selection depends on whether the current sub-block includes DC coefficients and whether any coeff_abs_level_greater1_flag equal to one has been transmitted for adjacent sub-blocks. for coeff_abs_level_greater1_flag, the context selection further depends on the number and value of the coded coeff_abs_level_greater1_flag's of the sub-block.

The sign coeff_sign_flag and the remainder of the absolute value coeff_abs_level_remaining are coded in the bypass mode of the binary arithmetic code writer. In order to map coeff_abs_level_remaining to the sequence of bits (binary decision), an adaptive binarization scheme is used. Binaryization is controlled by a single parameter, which is adjusted based on the coded value of the sub-block.

H.265| MPEG-H HEVC also includes the so-called symbol data hiding mode. In symbol data hiding mode (under certain conditions), the transmission of symbols used for the last non-zero level within the sub-block is omitted. Alternatively, the symbol for this level is embedded in the same bit of the sum of the absolute values of the level of the corresponding sub-block. It should be noted that the encoder must consider this aspect when determining the appropriate transform coefficient level.

Video coding standards only specify bit stream syntax and reconstruction procedures. If considering the transformation and coding of a given block and given quantization step size of the original prediction error sample, the encoder has a higher degree of freedom. Given the quantization index q _k of the transform block, the entropy writing code must follow a unique defining algorithm for writing data to the bit stream (ie, constructing an arithmetic codeword). However, the encoder algorithm used to obtain the quantization index q _k for the original block of the given prediction error sample is outside the scope of the video coding standard. In addition, the encoder can freely select the quantization parameter QP based on the block. For the following description, it is assumed that the quantization parameter QP and the quantization weight matrix are given. Therefore, the quantization step size of each transform coefficient is known. It is further assumed that the encoder performs an analytical transformation, which is an inverse transformation (or a very close approximation of the inverse transformation) of the designated synthetic transformation used to obtain the original transformation coefficient t _k. Even under these conditions, for each encoder is also capable of original transform coefficient t _k freely chosen quantizer index q _k. Since the choice of transform coefficient level determines both distortion (or reconstruction/approximation quality) and bit rate, the quantization algorithm used has a substantial impact on the rate-distortion performance of the generated bit stream.

其中sgn()為符號函數，且運算子

．

返回小於或等於其自變量之最大整數。此量化方法保證MSE失真

The simplest quantization method rounds the original transform coefficient t _k to the nearest reconstruction level. For the commonly used URQ, the corresponding quantization index q _{k can be determined according to the following formula}

Where sgn() is a symbolic function, and the operator

.

Returns the largest integer less than or equal to its argument. This quantization method guarantees MSE distortion

其中0

a<

. It is minimized, but it completely ignores the bit rate required for the transmission of the transform coefficient level q _k. Generally, better results are obtained if the rounding is biased towards zero:

Where 0

a <

.

之編 解碼器(諸如H.264 | MPEG-4 AVC或H.265 | MPEG-H HEVC)，常常使用拉格朗日乘數λ與量化參數QP之間的以下關係式

其中c ₁及c ₂表示圖塊或圖像之常數因子。 If the quantization program makes the Lagrangian function D + λ . R is minimized to obtain the best result in rate-distortion sensing, where D represents the distortion of the transform block (for example, MSE distortion), R specifies the number of bits required for the transform coefficient level of the transmission block, and λ is the Lagrangian multiplier. For the relationship between QP and quantization step size

The codec (such as H.264 | MPEG-4 AVC or H.265 | MPEG-H HEVC) often uses the following relationship between the Lagrangian multiplier λ and the quantization parameter QP

Where c ₁ and c ₂ represent the constant factors of the tiles or images.

Aim to make the Lagrangian function of distortion and rate D + λ . The quantization algorithm for minimizing R is also called rate-distortion optimized quantization (RDOQ). If MSE or weighted MSE is used to measure distortion, the quantization index q _{k of the} transformed block should be determined in a way that minimizes the following cost metrics:

Here, the transform coefficient index k specifies the coding order (or scanning order) of the transform coefficient hierarchy. The term R ( q _k | q _{k -1} , q _{k -2} , ...) represents the number of bits (or its estimated value) required to transmit the quantization index q _k. Conditional description (due to the use of combination or conditional probability), the number of bits in a specific transform coefficient level q _k usually depends on the previous transform coefficient levels q _{k -1} , q _{k -2,} etc. in the coding order Choose value. The factor α _k in the above formula can be used to weight the contribution of individual transform coefficients, for example, to model the contrast sensitivity of human vision. In the following, it is generally assumed that all weighting factors α _{k are} equal to 1 (but the algorithm can be directly modified by considering different weighting factors).

For the coding of transform coefficients in H.265|MPEG-H HEVC, the accurate calculation of the rate term is very complicated. This is because most of the binary The bit decision system uses an adaptive probability model to write codes. However, if some aspects of the probability model selection are ignored and the probability model is adjusted within the transformation block, it is possible to design an RDOQ algorithm with reasonable complexity. The RDOQ algorithm implemented with H.265|MPEG-H HEVC reference software consists of the following basic processing steps:

1. For each scanning position k , make the Lagrangian cost D _k ( q _k ) + λ under the assumption that the level is not inferred to be equal to zero. R _k ( q _k ) is minimized to select the transform coefficient level q _k . D _k ( q _k ) represents the weighted mean square error D _k ( q _k ) = α _k . ( t _k- △ _k . q _k ) ² , and R _k ( q _k ) represents the estimated value of the number of bits required to transmit q _k.

2. Determine the coded_sub_block_flag of the 4×4 sub-block by comparing the Lagrangian cost of the following two conditions: (a) Use the transform coefficient level selected in step 1; (b) Coded_sub_block_flag the syntax element It is set equal to zero, and therefore all transform coefficient levels of the 4×4 sub-blocks are set equal to zero.

3. Determine the position of the first non-zero transform coefficient level by comparing the Lagrangian cost. The Lagrangian cost is determined by selecting one of the non-zero transform coefficient levels (after step 2) as the coding order The first non-zero transform coefficient level in (set the previous transform coefficient level to be equal to zero).

4. Determine the coded_block_flag by comparing the Lagrangian cost of the sequence of transform coefficient levels obtained after step 3 and the condition that all transform coefficient levels inside the transform block are set equal to zero.

In [3], a modified concept for transform coding is proposed, which is described in more detail below. The main changes compared to the conventional coding The transform coefficients are not quantized and reconstructed non-dependently. Alternatively, the allowable reconstruction level of transform coefficients depends on the quantization index selected for the preceding transform coefficient in the reconstruction order. The concept of dependent scalar quantization is combined with modified entropy coding, where the probability model selection (or alternatively, the codeword table selection) of the transform coefficients depends on the set of allowable reconstruction levels.

之可容許值之集合不取決於第一經重構變換係數

之所選值。圖5b展示相依純量量化之實例。應注意，相比於非相依純量量化，第二變換係數

之可選擇重構值取決於第一變換係數

之所選重構層級。在圖5b之實施例中，針對第二變換係數

存在二個不同可用重構層級集合(由不同色彩說明)。若第一變換係數

之量化索引為偶數(…,-2,0,2,...)，則可針對第二變換係數

選擇第一集合(藍色點)之任何重構層級。且若第一變換係數

之量化索引為奇數(...,-3,-1,1,3,...)，則可針對第二變換係數

選擇第二集合(紅色點)之任何重構層級。在實例中，將第一集合及第二集合之重構層級移位一半量化步長(第二集合之任何重構層級位於第一集合之二個重構層級之間)。 The advantage of dependent quantization of transform coefficients is that it allows the reconstructed vector to be more densely packed in the N -dimensional signal space (where N represents the number of samples or transform coefficients in the transform block). The reconstruction vector of the transformation block refers to the ordered reconstruction transform coefficient of the transformation block (or alternatively, the ordered reconstruction sample). Figures 5a and 5b illustrate this effect for the simplest case of two transform coefficients. Figure 5a shows the allowable reconstruction vector (which represents a point in the 2d plane) for non-dependent scalar quantization. As can be seen, the second transform coefficient

The set of allowable values does not depend on the first reconstructed transform coefficient

The selected value. Figure 5b shows an example of dependent scalar quantification. It should be noted that compared to non-dependent scalar quantization, the second transform coefficient

The selectable reconstruction value depends on the first transform coefficient

The selected reconstruction level. In the embodiment of Figure 5b, for the second transform coefficient

There are two different sets of available reconstruction levels (illustrated by different colors). If the first transform coefficient

If the quantization index is an even number (...,-2,0,2,...), it can be used for the second transform coefficient

Select any reconstruction level of the first set (blue dots). And if the first transform coefficient

If the quantization index is an odd number (...,-3,-1,1,3,...), it can be used for the second transform coefficient

Select any reconstruction level of the second set (red dots). In the example, the reconstruction levels of the first set and the second set are shifted by half the quantization step (any reconstruction level of the second set is located between the two reconstruction levels of the first set).

The dependent scalar quantization of transform coefficients has the following effect: For a given average number of reconstruction vectors of each N -dimensional unit volume, the expected value of the distance between the given input vector of the transform coefficient and the closest available reconstruction vector is reduced small. Therefore, for a given average number of bits, the average distortion between the input vector of transform coefficients and the vector of reconstructed transform coefficients can be reduced. In vector quantization, this effect is called space filling gain. In the case of using dependent scalar quantization for transform blocks, the latent space of high-dimensional vector quantization can be used to fill the main part of the gain. Moreover, compared to vector quantization, the implementation complexity of the reconstruction process (or the decoding process) is equivalent to the complexity of coding with the conventional transformation using a non-dependent scalar quantizer.

不僅取決於相關聯量化索引q _k ，且亦取決於重構次序中之先前變換係數的量化索引q ₀ ,q ₁ ,…,q _k-1。應注意，在相依量化中，必須唯一地定義變換係數之重構次序。若在熵寫碼中亦利用關於與量化索引q _k 相關聯之重構層級集合的知識，則可改良總體變換編解碼器之效能。彼意謂基於適用於變換係數之重構層級集合而切換上下文(機率模型)或碼字表係有利的。 Figure 6 illustrates the block diagram of a transform decoder using dependent scalar quantization. The main change is about the dependent quantification of arrows that lead from top to bottom. As indicated by the vertical arrow, in the case of reconstruction order index k >0, the reconstructed transform coefficient

It not only depends on the associated quantization index q _k , but also depends on the quantization index q ₀ , q ₁ , ... , q _{k -1} of the previous transform coefficients in the reconstruction order. It should be noted that in dependent quantization, the reconstruction order of transform coefficients must be uniquely defined. If the knowledge about the reconstruction level set associated with the quantization index q _k is also used in the entropy coding, the performance of the overall transform codec can be improved. This means that it is advantageous to switch the context (probability model) or codeword table based on the set of reconstruction levels applicable to the transform coefficients.

As in the conventional transform coding, in addition to the analysis transform, the transform coding according to the embodiments outlined in this article also involves a quantization algorithm and entropy coding. The inverse transform of the synthetic transform (or the close approximation of the inverse transform) is usually used as the analysis transform, and in the case of a given entropy decoding program, it is usually unique Specify the entropy to write the code. However, similar to the conventional transform coding, given the original transform coefficients, there is a higher degree of freedom for selecting the quantization index.

The dependent quantization of transform coefficients refers to the following concept: the set of available reconstruction levels of transform coefficients depends on the quantization index selected for the previous transform coefficient in the reconstruction order (within the same transform block). Select the set of allowable reconstruction levels of the current transform coefficient from the two predefined sets of reconstruction levels (based on the quantization index of the previous transform coefficient in the coding order).

Similar to the conventional non-dependent scalar quantization, the quantization step size △ (or the corresponding scale and shift parameter as described above) is determined based on the quantization parameter (QP) of the block, and all reconstruction levels (in all reconstructions) In the level set) represents an integer multiple of the quantization step size △. However, each reconstruction level set only includes a subset of integer multiples of the quantization step △. This configuration in which all possible reconstruction levels in the set of all reconstruction levels represent integer multiples of the quantization step size can be regarded as an extension of the uniform reconstruction quantizer (URQ). The basic advantage is that the reconstructed transform coefficients can be calculated by an algorithm with extremely low computational complexity.

Not only can specific quantization parameter QP determination block transform coefficients by the quantization step size T _k △ _k (where k indicates the order of reconstruction), and may also block by the quantization matrix and the quantization weighting parameter _k determines particular transform coefficients T The quantization step size △ _k . Generally, the quantization step size Δ _k of a specific transform coefficient t _k is given by the product of the weighting factor w _{k of the} transform coefficient t _k (specified by the quantization weighting matrix) and the block quantization step size Δ _block (specified by the block quantization parameter) Set, △ _k = w _k . △ _block .

之實際計算(重構層級之實際計算)可略微偏離理想乘法。使△ _k 為特定變換係數t _k 之量化步長且使n _k 指定量化步長之標稱整數因子(例如，由量化索引q _k 給定)。在理想乘法之情況下，經重構變換係數

由下式給定

It should be noted that due to integer implementation or other implementation aspects, the reconstructed transform coefficient

The actual calculation (reconstruction level actual calculation) may slightly deviate from the ideal multiplication. Let Δ _k be the quantization step size of the specific transform coefficient t _k and let n _k designate the nominal integer factor of the quantization step size (for example, given by the quantization index q _k ). In the case of ideal multiplication, the reconstructed transform coefficient

Given by

(或對應重構層級)實際上可根據下式

，其中scale．2^-shift

△ _k 。 Due to restrictions on the implementation of integers, reconstructed transform coefficients

(Or corresponding reconstruction level) can actually be based on the following formula

, Where scale. 2 ^-shift

△ _k .

Or similar procedures for judgment. If an integer multiple of the quantization step is mentioned in the following description, the corresponding text also applies to an integer approximation similar to the approximation specified above.

The dependent scalar quantization of transform coefficients proposed in [3] uses two different sets of reconstruction levels, and all reconstruction levels of the two sets of _{transform coefficients t k} represent integers _{of the quantization step △ k of this transform coefficient} Times (which is determined at least in part by block-based quantization parameters). It is noted that the quantization step △ _k shows only two of the set of reconstructed value of allowable ratio factor. In addition to the different transform blocks of transformation coefficient t may be inside individual outside (and hence the individual scale factors), the same two-level set of reconstructed quantization step △ _{k k} for all of the transform coefficients.

In FIG. 7, the preferred configuration of the two reconstruction level sets t is illustrated. The reconstruction level contained in the first quantization set (marked as set 0 in the figure) represents an even integer multiple of the quantization step. The second quantization set (marked as set 1 in the diagram) contains all odd integer multiples of the quantization step and additionally contains Reconstruction level equal to zero. It should be noted that the two reconstruction sets are symmetric around zero. The two reconstruction sets contain a reconstruction level equal to zero, otherwise the reconstruction sets do not intersect. The union of the two reconstruction sets contains all integer multiples of the quantization step.

The reconstruction level selected by the encoder among the allowable reconstruction levels is transmitted within the bit stream. As in the conventional non-dependent scalar quantization, this can be achieved by using so-called quantization indexes, which are also called transform coefficient levels. The quantization index (or transform coefficient level) is an integer that uniquely identifies the available reconstruction level within the quantization set (that is, the reconstruction level set). The quantization index is sent to the decoder as part of the bit stream (using any entropy coding technique). On the decoder side, the reconstructed transform coefficient can be uniquely calculated based on the current set of reconstruction levels (which is determined by the previous quantization index in the coding/reconstruction order) and the transmitted quantization index of the current transform coefficient.

The assignment of the quantization index to the reconstruction level inside the reconstruction level set (or quantization set) can follow the following rule as can be seen in FIG. 7: the quantization index equal to 0 is assigned to the reconstruction level equal to 0. Assign a quantization index equal to 1 to the smallest reconstruction level greater than 0, assign a quantization index equal to 2 to a reconstruction level greater than 0 (ie, the second smallest reconstruction level greater than 0), and so on. Or, in other words, the reconstruction levels greater than 0 are marked with integers greater than 0 (that is, 1, 2, 3, etc.) in the increasing order of their values. Similarly, the quantization index -1 is assigned to the largest reconstruction level less than 0, the quantization index -2 is assigned to the reconstruction level below 0 (ie, the second largest), and so on. Or, in other words, the reconstruction levels less than 0 are marked with integers less than 0 (ie, -1, -2, -3, etc.) in the descending order of their values. The reconstruction procedure of the transform coefficients can be implemented similarly to the algorithm specified in the pseudo code of FIG. 8.

In the pseudo code of Figure 8, level[k] represents the quantization index transmitted for the transform coefficient t _k , and setId[k] (equal to 0 or 1) specifies the identifier of the current reconstruction level set (which is based on the reconstruction The previous quantization index in the order is determined, as will be described in more detail below). The variable n represents an integer multiple of the quantization step given by the quantization index level[k] and the set identifier setId[k]. When reconstituted using the first level of the coupling comprises an integral multiple of the quantization step △ _k of the set of transform coefficient code write (setId [k] == 0) , the variable n is the index of the transmitted quantization twice. If the second reconstruction level set is used to write the transform coefficients (setId[k]==1), there are the following three conditions: (a) If level[k] is equal to 0, then n is also equal to 0; (b) ) If level[k] is greater than 0, then n is equal to twice the quantization index level[k] minus 1; and (c) if level[k] is less than 0, then n is equal to twice the quantization index level[k] plus 1. This can be specified using the following symbolic functions:

Then, if the second quantization set is used, the variable n is equal to twice the quantization index level[k] minus the sign function sign(level[k]) of the quantization index.

。 Once the variable n (an integer factor specifying the quantization step size) is determined, the reconstructed transform coefficient is obtained by multiplying _{n and the quantization step size △ k}

.

可藉由整 數近似獲得，而非與量化步長△_k精確相乘。此說明於圖9中之偽碼中。此處，變數移位表示向右的位元移位。其值通常僅取決於區塊之量化參數(但移位參數可針對區塊內部之不同變換係數改變亦係可能的)。變數scale[k]表示變換係數t _k 之縮放因子；除區塊量化參數以外，其亦可例如取決於量化加權矩陣之對應輸入項。變數「add」指定捨入偏移，其通常被設定為等於add=(1<<(shift-1))。應注意，除了捨入之外，圖9之偽碼(最末行)中所之指定的整數算術等效於與由下式給定之量化步長△_k相乘△_k=scale[k]．2^-shift. As mentioned above, the reconstructed transform coefficients

May be approximated by an integer, instead of the quantization step size is multiplied with the exact △ _k. This is illustrated in the pseudo code in Figure 9. Here, variable shift means bit shift to the right. Its value usually depends only on the quantization parameter of the block (but it is also possible that the shift parameter can be changed for different transform coefficients within the block). The variable scale[k] represents the scaling factor of the transform coefficient t _k ; in addition to the block quantization parameter, it can also depend on the corresponding input item of the quantization weighting matrix, for example. The variable "add" specifies the rounding offset, which is usually set equal to add=(1<<(shift-1)). It should be noted that, except for rounding, the integer arithmetic specified in the pseudocode (the last row) of Figure 9 is equivalent to multiplying _{by the quantization step △ k given by the following formula △ k} =scale[ k ]. 2 ^-shift .

Another (appearance only) change in Figure 9 compared to Figure 8 is to use a ternary if-then-else operator (a?b:c) to switch between the two reconstruction level sets. This operator Known from programming languages such as the C programming language.

In addition to the selection of reconstruction level sets discussed above, another task in dependent scalar quantization in transform coding is an algorithm for switching between defined quantization sets (reconstruction level sets). The algorithm used determines the "packaging density" that can be achieved in the N -dimensional space of the transform coefficients (and therefore also in the N-dimensional space of the reconstructed samples). The higher packing density ultimately leads to improved coding efficiency.

In the concept presented in [3], the reconstruction level set of the next transform coefficient is determined based on the division of the quantized set, as illustrated in FIG. 10. Divide each of the two quantized sets into two subsets. The first quantized set (labeled as set 0) is divided into two subsets (labeled as A and B), and the second quantized set (labeled as set 1) is also divided into two subsets (labeled as C and D) ). In Figures 7 and 10, the division of the quantized set into subsets is indicated by open circles and solid circles. The following segmentation rules apply: ‧Subset A is composed of all the even quantization indexes of quantization set 0; ‧Subset B is composed of all the odd quantization indexes of quantization set 0; ‧Subset C is composed of all the even quantization indexes of quantization set 1 ‧Subset D is composed of all odd quantization indexes of quantization set 1; the used subset is not clearly indicated in the bit stream. Alternatively, it can be derived based on the quantization set used (for example, set 0 or set 1) and the quantization index actually transmitted. The subset can be derived from the bitwise "and" operation of the transmitted quantization index level and 1. Subset A is composed of all quantization indexes of set 0 where (level&1) is equal to 0, subset B is composed of all quantization indexes of set 0 where (level&1) is equal to 1, and subset C is composed of all quantization indexes of set 1 where (level&1) is equal to 0 It is composed of quantization indexes, and subset D is composed of all quantization indexes of set 1 whose (level&1) is equal to 1.

The transition between quantized sets (set 0 and set 1) is represented by a state variable; the state variable has four possible values (0, 1, 2, 3). On the one hand, the state variable specifies the quantized set for the current transform coefficient. If and only if the state variable is equal to 0 or 1, the quantization set 0 is used, and if and only if the state variable is equal to 2 or 3, the quantization set 1 is used. On the other hand, state variables also specify possible transitions between quantized sets. Table 1 shows the state transition table used. Given the current state, it specifies the quantized set of current transform coefficients (the second line). It further specifies a state transition based on the path associated with the selected quantization index (if the quantization set is given, the path Specify the subset A, B, C, or D used). When reconstructing the transform coefficients of the block, it is sufficient to update the state variables and determine the path of the quantization index used.

The path is given by the parity of the quantization index. In the case that level[k] is the current quantization index, it can be determined according to the following formula path=(level[k]&1), where the operator & represents the bitwise “and” in two's complement integer arithmetic.

Using the concept of state transition, the current state and therefore the current quantization set are uniquely determined by the previous state (in reconstruction order) and the previous quantization index, such as its parity in this example. The first state of the transformation block is set equal to 0.

The concept of state transition for dependent scalar quantization allows low-complexity implementation for reconstructing transform coefficients in the decoder. A preferred example of the reconstruction procedure of the transform coefficients of a single transform block is shown in FIG. 11 using C-style pseudo code.

In the pseudo code of FIG. 11, the index k specifies the reconstruction order of transform coefficients. It should be noted that in the example code, the index k decreases in reconstruction order. The last transform coefficient has an index equal to k =0. The first index kstart specifies the reconstruction index (or more accurately, the inverse reconstruction index) of the first reconstructed transform coefficient. The variable kstart can be set equal to the number of transform coefficients in the transform block minus 1, or it can be set equal to the index of the first non-zero quantization index in the coding/reconstruction order (for example, if writing with the applied entropy The coding method transmits the position of the first non-zero quantization index). In the latter case, it is inferred that all previous transform coefficients (where index k > kstart ) are equal to zero. The reconstruction procedure of each single transform coefficient is the same as the reconstruction procedure in the example of FIG. 9. For the example in FIG. 9, the quantization index is represented by level[k], and the associated reconstructed transform is represented by trec[k]. State variables are represented by states. It should be noted that in the example of Figure 11, the state is set equal to 0 at the beginning of the transform block. The 1d table setId[] specifies the quantization set associated with the different values of the state variables, and the 2d table state_trans_table[][] specifies Given the current state (the first argument) and the state transition of the path (the second argument). The path is given by the parity of the quantization index (using the bitwise and operator &).

Simple arithmetic operations can be used instead of the table setId[]; for example, bit shift to the right: setId[state]=state>>1

Similarly, simple arithmetic operations can also be used to implement the table state_trans_table[][]. For example, state_trans_table[state][path]=(32040>>((state<<2)+(path<<1)))&3

Here, the complete state transition table is given by the 16-bit value "32040". It should be noted that by replacing the value "32040" with "0", it is easy to switch between dependent quantization and non-dependent quantization. The value "0" means that the state 0 is always selected and therefore the state transition table of the conventional uniform reconstruction quantizer Q0 is selected.

The state transition in dependent quantization can also be represented by a grid structure, as illustrated in Figure 12. The grid shown in this figure corresponds to the state transitions specified in Table 1. For each state, there are two paths connecting the state of the current transform coefficient and the two possible states of the next transform coefficient in the reconstruction order. The paths are labeled with path 0 and path 1, and this number corresponds to the path variable introduced above (for the preferred embodiment where the path variable is equal to the quantization index). It should be noted that each path uniquely specifies a subset (A, B, C, or D) of the quantization index. In Figure 12, the subset is specified in parentheses. Given the initial state (state 0), the transmitted quantization index uniquely specifies the path through the grid.

For example, in Figure 12, the states (0, 1, 2, and 3) have the following properties:

‧State 0 : The previous quantization index level[k-1] specifies the reconstruction level of set 0, and the current quantitative index level[k] specifies the reconstruction level of set 0.

‧State 1 : The previous quantitative index level[k-1] specifies the reconstruction level of set 1, and the current quantitative index level[k] specifies the reconstruction level of set 0.

‧State 2 : The previous quantitative index level[k-1] specifies the reconstruction level of set 0, and the current quantitative index level[k] specifies the reconstruction level of set 1.

‧State 0 : The previous quantization index level[k-1] specifies the reconstruction level of set 1, and the current quantitative index level[k] specifies the reconstruction level of set 1.

The grid consists of a series of so-called basic grid cells. The basic grid cell is shown in Figure 13.

As outlined above, the aspect of dependent scalar quantization is that there are different sets of allowable reconstruction levels for transform coefficients (also known as quantitative 化集). The quantization set of the current transform coefficient is determined based on the value of the quantization index of the previous transform coefficient. Comparing the two quantized sets, it is obvious that in set 0, the distance between the reconstruction level equal to zero and the adjacent reconstruction level is greater than in the case of set 1. Therefore, if set 0 is used, the probability that the quantization index is equal to 0 is greater, and if set 1, the probability is smaller. For efficient coding, this aspect is used for entropy coding by switching probability models based on the quantization set (or state) used for the current quantization index.

It should be noted that in order to properly switch the probability model, when entropy decoding the current quantization index (or the corresponding binary decision of the current quantization index), all the paths of the previous quantization index (associated with the subset of the used quantization set) must A known. For that purpose, in the codec proposed in [3], the transform coefficients are written in reconstruction order. Use binary arithmetic coding code similar to H.264|MPEG-4 AVC or H.265|MPEG-H HEVC to code the quantization index. The non-binary quantization index is first mapped to a series of binary decisions (which are usually referred to as bits). The quantization index is transmitted as an absolute value and as a symbol when the absolute value is greater than 0.

Similarly, for HEVC, the transform coefficient level of the transform block is coded based on the sub-block. First, the transmission flag coded_block_flag, which specifies whether the transform block contains any non-zero transform coefficient levels. If coded_block_flag is equal to 0 (that is, the block does not contain any non-zero levels), no additional information is transmitted for the transformed block. Otherwise (coded_block_flag is equal to 1), the following applies:

‧The x and y coordinates of the first non-zero level in the coding sequence. As shown in Figure 14 As explained in, the transmitted position designation of the first non-zero level is inferred that all transform coefficients (marked in white in FIG. 14) that precede the identified coefficients in the coding order are equal to zero. Additional data is transmitted only for the coefficient at the designated position (marked in black in FIG. 14) and the coefficient after this coefficient in the coding order (marked by hatching in FIG. 14). The example in Figure 14 shows a 16×16 transform block with 4×4 sub-blocks; the coding order used is the sub-block diagonal scan specified in H.265|MPEG-H HEVC.

‧For the sub-block that precedes the other sub-block that contains the first non-zero level (indicated by the transmitted x and y coordinates), the transmission flag coded_subblock_flag, which specifies whether the sub-block contains any non-zero transform coefficient levels. As an exception, coded_subblock_flag is not transmitted for sub-blocks containing DC coefficients. For this subblock, it is inferred that coded_subblock_flag is equal to 1.

Finally, for all sub-blocks with coded_subblock_flag equal to 1 and sub-blocks containing the first non-zero level in the coding order, the value of the transform coefficient level is coded, as illustrated by the pseudo code in FIG. 15. Here, firstScanIdSbb represents the first scan index inside the sub-block. For the sub-block containing the first non-zero level (indicated by the transmitted x and y coordinates), firstScanIdSbb is equal to the scan index firstNonZero corresponding to the transmitted (x, y) coordinates. For other transmitted sub-blocks, firstScanIdSbb specifies the first scan index within the sub-block. The scan index lastScanIdSbb specifies the last scan index within the sub-block. It should be noted that all coefficients in the sub-block are written before any coefficients of the next sub-block in the coding order are written.

In the second pass, the level of the sub-block is coded. The absolute value is transmitted in the first pass, and the symbol is transmitted for all coefficients whose absolute value is not equal to zero in the second pass. Write the absolute value as follows:

‧Transmission flag sig_flag, which specifies whether the absolute level is greater than zero. If it can be inferred that the flag is equal to 1, that is, this flag is not transmitted when any of the following conditions apply: ○ The current scan index k is equal to the first non-zero level (as indicated by the transmitted x and y coordinates ) Scan index; ○ The current scan index is the last scan index inside the sub-block, coded_subblock_flag equal to 1 has been transmitted for the sub-block, and all previous levels in the sub-block are equal to 0.

‧If sig_flag is equal to 1 (that is, the absolute level is greater than 0), then another flag gt1_flag is transmitted, which specifies whether the absolute level is greater than one.

‧If sig_flag and gt1_flag are equal to 1 (that is, the absolute level is greater than 1), another flag gt2_flag is transmitted, which specifies whether the absolute level is greater than two.

‧If sig_flag, gt1_flag and gt2_flag are equal to 1 (that is, the absolute level is greater than 2), then another flag gt3_flag is transmitted, which specifies whether the absolute level is greater than three.

‧If sig_flag, gt1_flag, gt2_flag, and gt3_flag are equal to 1 (that is, the absolute level is greater than 3), then another flag gt4_flag is transmitted, which specifies whether the absolute level is greater than four.

‧If sig_flag, gt1_flag, gt2_flag, gt3_flag, and gt4_flag are equal to 1 (that is, the absolute level is greater than 4), the syntax element is transmitted The prime gt3_flag, which specifies the absolute value minus 5.

After transmitting the absolute values of all levels of the sub-block, the sign bit sign_flag is transmitted for all coefficients whose absolute level is not equal to zero.

Use an adaptive probability model (also called context) to write the flags sig_flag, gt1_flag, gt2_flag, gt3_flag, and gt4_flag. For these flags, one of multiple adaptive probability models is selected as follows: the state is the current value of the state variable of the transformation coefficient (the state variable is determined based on the parity of the code level in the coding sequence). Make diag specify the diagonal position of the current scan position (diag=x+y, where x and y specify the x and y coordinates of the scan position). In addition, let sumAbsTemplate be the sum of the absolute values of the written codes in the local neighborhood illustrated in Figure 16, where the corresponding neighborhood is indicated by shading the coefficients belonging to the neighborhood of the coefficients. Yuanzi is currently writing the code, and the position is again shown in black. And let numSigTemplate be the absolute number of levels greater than 0 in the same local neighborhood.

For sig_flag, the selected probability model depends on: ‧The diagonal of the current scan position (given by the sum of x and y coordinates); ‧The quantizer used for the current transform coefficient (Q0 or Q1, by state>>1 Given); ‧The sum of the absolute levels of code written in the local neighborhood given by sumAbsTemplate.

For gt1_flag, the selected probability model depends on: ‧The diagonal of the current scan position (given by the sum of x and y coordinates); ‧Quantizer used for the current transform coefficient (Q0 or Q1, given by state>>1); ‧Value sumAbsTemplate-numSigTemplate (that is, the sum of the absolute values in the local neighborhood minus the value greater than 0 in the local neighborhood Absolute number of levels).

For gt2_flag, gt3_flag, and gt4_flag, the same probability model is used. It depends on the following to choose: ‧The diagonal of the current scan position (given by the sum of the x and y coordinates); ‧The value sumAbsTemplate-numSigTemplate (that is, the sum of the absolute values in the local neighborhood minus the local The absolute number of levels in the neighborhood greater than 0).

First, the non-binary syntax elements are binarized (that is, mapped to the sequence of bit subs), and in the bypass mode of the arithmetic coding engine (using the non-adaptive probability model with pmf{0.5,0.5} ) Write code to the bit. For binarization, Rice-Golomb codes are used, which are parameterized by so-called Rice parameters. The Rice parameter is selected depending on the value of (sumAbsTemplate-numSigTemplate). Finally, use the bypass mode of the arithmetic coding engine to code the symbol flag.

In order to obtain a bit stream that provides an excellent trade-off between distortion (reconstruction quality) and bit rate, the Lagrangian cost metric should be

且因此其失真

不僅取決於相關聯量化索引q _k ，且亦取決於寫碼次序中之先前量化索引之值。 Select the quantization index in a minimized way. For non-dependent scalar quantization, this quantization algorithm (called rate-distortion optimized quantization or RDOQ) is discussed above. But compared with non-dependent scalar quantification, it has additional difficulties. Reconstructed transform coefficient

And therefore its distortion

It depends not only on the associated quantization index q _k , but also on the value of the previous quantization index in the coding order.

However, as discussed above, the dependency between transform coefficients can be expressed using a grid structure. For further description, we use the preferred embodiment given in FIG. 10 as an example. The grid structure of an example of blocks of 8 transform coefficients is shown in FIG. 17. The path through the grid (from left to right) represents the possible state transitions of the quantization index. It should be noted that each connection between two nodes represents the quantization index of a specific subset (A, B, C, D). If the quantization index q _k is selected from each of the subsets (A, B, C, D) and the corresponding rate-distortion cost J _k = D _k ( q _k | q _{k -1} ,q _{k -2} , …) + λ ． R _k ( q _k | q _{k -1} ,q _{k -2} , …)

Assigned to the associated connection between two grid nodes, it is determined to make the total rate-distortion cost D + λ . The problem of the vector/block of the quantization index for minimizing R is equivalent to finding the path with the least cost path through the grid (from left to right in Figure 17). If some dependencies of entropy coding (similar to RDOQ) are ignored, the well-known Viterbi algorithm can be used to solve this minimization problem.

The example coding algorithm used to select the appropriate quantization index of the transform block can consist of the following main steps:

1. Set the initial state rate-distortion cost equal to zero.

2. For all transform coefficients in the coding sequence, perform the following operations:

a. For each subset A, B, C, and D, determine the quantization index that minimizes the distortion of the given original transform coefficient.

b. For all grid nodes (0, 1, 2, 3) of the current transformation coefficient, perform the following operations:

i. Calculate the rate-distortion cost of the two paths connecting the state of the previous transform coefficient and the current state. The cost is given as the cost of the previous state and D _k + λ . The sum of R _k , where D _k and R _k represent the distortion and rate of the quantization index used to select the subset (A, B, C, D) associated with the connection under consideration.

ii. Assign the minimum value of the calculated cost to the current node and trim the connection to a state that does not represent the previous transformation coefficient of the least cost path.

It should be noted that after this step, all nodes of the current transform coefficient have a single connection to any node of the previous transform coefficient.

3. Compare the cost of the 4 final nodes (for the last coefficient in the coding order) and select the node with the smallest cost. It should be noted that this node is associated with a unique path through the grid (all other connections are trimmed in the previous step).

4. Follow the selected path (specified by the final node) in reverse order and collect the quantized index associated with the connection between the grid nodes.

Now, embodiments of the present application will be described. It is presented in a standalone manner, but sometimes with reference to the schema discussed above. In particular, the following description focuses on the differences from the above examples, and therefore, these differences can be used to modify the above description to obtain other embodiments, and vice versa, the individual tasks described above can be used individually or in combination , Such as about context selection Select, quantize, dequantize, transform, retransform, entropy coding/decoding tasks to modify or further specify the subsequently explained embodiments to obtain even other embodiments.

Compared with conventional code writing using non-dependent scalar quantization, the conversion code using dependent scalar quantization [3] as described above generally improves coding efficiency. However, the entropy coding of the dependent quantization index (transform coefficient level) described in Figure 15 has two problems, which make it difficult to achieve a high-throughput hardware design:

‧Bits (sig_flag, gt1_flag, gt2_flag, gt3_flag and gt4_flag) for writing codes in the regular coding mode of the arithmetic coding engine and for writing codes in bypass mode (bits of the remainder of the syntax element) staggered. The bypass mode of the arithmetic coding engine can be implemented more efficiently than the conventional mode. And in addition, when a relatively large number of bypass bits are continuously written, the bypass mode of arithmetic coding can be implemented particularly efficiently. Frequent switching between normal mode and bypass mode is not conducive to hardware implementation.

‧The probability model of most bits directly depends on the value of the bits directly in front. For example, the probability model used for sig_flag depends on the parity of the previous absolute level (which is unknown before the last bit of the previous level is read). Then, the value of sig_flag determines whether the next bit is gt1_flag or another sig_flag. If gt1_flag is read, its value determines whether the next bit is gt2_flag or sig_flag. These direct dependencies prevent efficient pipelineization of arithmetic decoding procedures. It would be better to reduce the number of direct dependencies so that the arithmetic decoding of bits can be pipelined to a certain extent.

Reduce the complexity of the current advanced technology residual coding design The method is to write a code in several scan passes. In each scan pass, part of the level information is transmitted so that the complete level information is only available after the final scan pass. However, this design is not compatible with dependent quantization techniques. It should be noted that the reordering of bits is not simple for dependent quantification due to the following reasons:

‧In particular, for the flag sig_flag, there are significantly different probability quality functions for the two supporting quantizers. If the quantizer (Q0 or Q1) of the current transform coefficient is not known, the compression efficiency will be significantly degraded. But the quantizer used depends on the parity of all transform coefficient levels. The parity is only when all the bits of all previous transform coefficients are written before the first bit (sig_flag) of the current transform coefficient. known.

‧The contextual modelling of bit sons also depends on the knowledge of the absolute level that has been transmitted in the local neighborhood. Generally, better knowledge about adjacent transform coefficient levels improves coding efficiency.

Frankly speaking, the embodiments described further below overcome the problems mentioned by one or more of the following design aspects:

‧Write the bits related to the transformation coefficient level of the sub-block (or block) in multiple passes over all scanning positions.

‧The first pass includes the writing code of sig_flag in one or more scanning positions, and if sig_flag is equal to 1, it includes the code writing of the parity flag par_flag (the parity of the specified transform coefficient level). It may or may not include additional information. It should be noted that with the dedicated parity flag for coding, the quantizer used for transform coefficients is known, and this knowledge can be used for efficient context modeling of sig_flag's.

‧Write all the bypass bits of the remainder of the syntax element of the sub-block (or block) in a single pass. This means that all bypass bits of the sub-block (or block) are written continuously.

In the following, a design method is described, which is firstly compatible with quantization methods that require co-location information, and secondly, it inherits a lower complexity than the coding of complete absolute-level information. The central concept is to transmit the parity information of each absolute level as a special syntax element. Different positions are possible, such as before the effective value flag, after the effective value flag, and after any "level greater than..." information. The semantics of the syntax elements that follow the parity information alternate depending on the selected position for transmitting the parity syntax elements. When parity information is available, the remaining absolute level information can be divided by two to generate different conditional probabilities for syntax elements.

The following describes other details about the coding order of bit elements and associated context modeling.

In the embodiment, similar to HEVC, the transform coefficient level of the transform block is coded based on sub-blocks. First, the transmission flag coded_block_flag, which specifies whether the transform block contains any non-zero transform coefficient levels. If coded_block_flag is equal to 0 (that is, the block does not contain any non-zero levels), no additional information is transmitted for the transformed block. Otherwise (coded_block_flag is equal to 1), the following applies:

‧Transmit the x and y coordinates of the first non-zero level in the coding sequence. As illustrated in FIG. 14, the transmitted position designation of the first non-zero level is inferred that all transform coefficients (marked in white in FIG. 14) that precede the identified coefficients in the coding order are equal to zero.

‧For the sub-block that contains the first non-zero level in the coding sequence (indicated by the transmitted x and y coordinates), the transmission flag coded_subblock_flag, which specifies whether the sub-block contains any non-zero Zero transform coefficient level. As an exception, coded_subblock_flag is not transmitted for sub-blocks containing DC coefficients. For this subblock, it is inferred that coded_subblock_flag is equal to 1.

Finally, for all sub-blocks with coded_subblock_flag equal to 1 and sub-blocks containing the first non-zero level in the coding order, the value of the transform coefficient level is coded, as described below.

The preferred embodiment for coding the transform coefficient level of the sub-block is illustrated by the pseudo code in FIG. 18. Here, firstScanIdSbb represents the first scan index inside the sub-block. For the sub-block containing the first non-zero level (indicated by the transmitted x and y coordinates), firstScanIdSbb is equal to the scan index firstNonZero corresponding to the transmitted (x, y) coordinates. For other transmitted sub-blocks, firstScanIdSbb specifies the first scan index within the sub-block. The scan index lastScanIdSbb specifies the last scan index within the sub-block.

Let level[k] and absLevel[k]=abs(level[k]) represent the absolute value of the transform coefficient level and the transform coefficient level at the scanning position k. The coding of the transform coefficient level continues in four passes throughout the scan position inside the sub-block: In the first pass, the binary syntax elements sig_flag, par_flag, and gt1_flag are transmitted: ‧The binary syntax element sig_flag[k] specifies the change at scan position k Is the absolute value of the conversion factor level greater than 0, that is, sig_flag[k]=(absLevel[k]>0? 1:0).

If it can be inferred that sig_flag is equal to 1, that is, when any of the following conditions applies, sig_flag is not transmitted: ○ The current scan index k is equal to the first non-zero level (as indicated by the transmitted x and y coordinates) Scan index; ○ The current scan index is the last scan index within the sub-block, coded_subblock_flag equal to 1 has been transmitted for the sub-block, and all previous levels in the sub-block are equal to 0.

‧If sig_flag[k] is equal to 1, then the binary syntax elements par_flag[k] and gt1_flag[k] are transmitted.

par_flag[k] specifies the parity of transform coefficient levels. In a preferred embodiment of the present invention, par_flag[k] is set to the parity equal to the absolute value minus 1 (which represents the inverse parity of the transform coefficient level): par_flag[k]=(absLevel[k]-1 )&1.

gt1_flag[k] specifies whether the remainder (given by the value of sig_flag[k]=1 and par_flag[k]) is greater than zero: gt1_flag[k]=(((absLevel[k]-1)>>1)>0? 1: 0), where the operator ">>" specifies the bit shift to the right (that is, divisible by 2).

In the second pass, the binary syntax element gt2_flag is transmitted:. Write the binary syntax element gt2_flag[k] only for their scan position k, where the transmission is equal to 1 in the first pass gt1_flag[k]. gt2_flag[k] specifies whether the remainder (given by the value of sig_flag[k]=1 and par_flag[k]) is greater than one: gt2_flag[k]=(((absLevel[k]-1)>>1)>1? 1:0).

In the third pass, transfer the remainder of the syntax element:. Write code to the syntax element remainder[k] only for their scan position k, where gt2_flag[k] equal to 1 is transmitted in the second pass. remainder[k] specifies the remainder of the absolute value (given by the value of sig_flag[k]=1, gt1_flag[k]=1, gt2_flag[k]=1 and par_flag[k]): remainder[k]=((absLevel [k]-1)>>1)-2.

Finally, in the fourth pass, the syntax element sign_flag is transmitted: • Write the syntax element sign_flag[k] only for their scan position k, and in the first pass, it is inferred to transmit sig_flag[k] equal to 1. sign_flag[k] specifies whether the transform coefficient level is negative: sign_flag[k]=(level[k]<0? 1:0).

On the decoder side, the syntax elements sig_flag[k], par_flag[k], gt1_flag[k], gt2_flag[k], remainder[k], and sign_flag[k] are similarly decoded from the bit stream. It should be noted that all values of the syntax elements par_flag[k], gt1_flag[k], gt2_flag[k], remainder[k], and sign_flag[k] that have not been transmitted are inferred to be equal to zero. Under the condition that sig_flag[k] is not transmitted for the sub-block whose coded_subblock_flag is equal to 1, it is inferred that its value is equal to 1.

Given the written code and the inferred value, the absolute value of the transform coefficient level at the scan position k can be reconstructed as follows: absLevel[k]=sig_flag[k]+par_flag[k]+2*(gt1_flag[k]+gt2_flag[k]+remainder[k]).

And when the sign_flag[k] of the given absolute level is not equal to 0, the transform coefficient level level[k]=(sign_flag[k]?-absLevel[k]: absLevel[k]) is given by the following formula.

The specific embodiments described above can be modified in one or more aspects, for example:

‧The coding of transform coefficient level can not be based on sub-blocks. This means that all transform coefficient levels of a transform block can be coded at once, instead of splitting the coding of the transform coefficient level into sub-blocks. In this situation, the pass described above means the pass of all scanning positions inside the complete transformation block. This method can still be combined with the coding combination of coded_block_flag and the indication of the scanning position of the first non-zero level in the coding sequence (for example, by transmitting the x and y positions or by any other means). This method can also be combined with an indication (similar to coded_block_flag's) that all transform coefficient levels within the designated large area are equal to 0. These regions can represent sub-blocks of transform coefficient positions, continuous scan positions, or any other well-defined subsets. The corresponding indication can be coded before the actual transform coefficient level, or it can be coded alternately with the bits of the first pass.

‧The coding sequence of flag par_flag and gt1_flag can be changed. It should be noted that these flags do not depend on each other, and therefore, par_flag can be written before gt1_flag, or gt1_flag can be written before par_flag.

‧The meaning of par_flag can be changed. The parity of the absolute level (which is the same as the parity of the absolute level minus 2) can be transmitted instead of the parity of the absolute level minus 1.

‧The meaning of gt1_flag can be changed. It is also possible to transmit whether (absLevel-1-par_flag) is greater than 0, but not whether (absLevel-1)>>1 is greater than 0 (see above). Or, if the meaning par_flag is changed as indicated above, gt1_flag may indicate whether (absLevel-1) is greater than 0. In the two cases, the reconstruction formula will be changed to absLevel[k]=sig_flag[k]+par_flag[k]+gt1_flag[k]+2*(gt2_flag[k]+remainder[k])

And the meaning of gt2_flag and remainder will be changed to gt2_flag[k]=(((absLevel[k]-2)>>1)>0? 1:0)

remainder[k]=((absLevel[k]-2)>>1)-1

See, for example, Figure 19, where the flags are distributed in different ways on the pass, and the reconstruction of the quantization index involves absQIdx=sig_flag+gt1_flag+par_flag+2 *(gt3_flag+remainder), of which the first shown in Figure 19 The feasible part reconstruction after the pass is absQIdx1=sig_flag+gt1_flag+par_flag+2 * gt3_flag.

‧The first pass can only be modified by transmitting sig_flag and par_flag in this pass. The gt1_flag can be moved to the second pass, or it can be transmitted in a separate pass between the first pass and the second pass described. Alternatively, an additional flag (for example, gt2_flag) may be coded as a second part of the first pass.

‧The second pass (with gt2_flag's) can be omitted. In this case, the pass with the remainder of the syntax element will be written directly after the first pass. In one embodiment, the second pass is omitted because gt2_flag has already been coded as part of the first pass (see above). In another embodiment, the second pass is omitted because gt2_flag is not transmitted at all. In the latter case, the meaning of the remainder of the grammatical element is changed to remainder[k]=((absLevel[k]-1)>>1)-1

And the reconstruction type is changed to absLevel[k]=sig_flag[k]+par_flag[k]+2*(gt1_flag[k]+remainder[k])

Or, when the meaning of gt1_flag is changed as described above, the reconstruction formula is changed to absLevel[k]=sig_flag[k]+par_flag[k]+gt1_flag[k]+2*remainder[k]

Alternatively, the second pass may include additional flags. For example, the meaning is the following additional gt3_flag gt3_flag[k]=(((absLevel[k]-1)>>1)>2? 1:0)

Can be transmitted. Or, as mentioned above, gt1_flag can be moved from the first pass to the second pass. This can also be combined with gt3_flag.

It is also possible to adaptively determine the maximum number of gtx_flags transmitted for the current scanning position based on the coded data (for example, based on the sum of the absolute values in the local neighborhood derived from the coded data) Item.

‧One or more additional passes with regular writing code positions can write code between the second pass (with gt2_flag's) and the third pass (syntax element remainder). For example, gt3_flag's can be transmitted in additional passes.

‧The bypass code bits and the bypass code sign_flag's of the remainder of the syntax element can be written alternately in a single pass.

It is also possible to combine two or more points listed above.

The above embodiments are briefly summarized, and in the following, other embodiments are also described. In this way, reference symbols that point to the above-discussed diagrams are used.

In particular, the above embodiments describe the concept for decoding the block (10) of transform coefficients 12, in which the transform block 10 may or may not be subdivided into sub-blocks 14.

Decoding is performed in passes. In one or more first passes of scan transform coefficients, decode the effective value flag of the current transform coefficient indicating whether the quantization index of the transform coefficient is zero, and decode the parity flag of the transform coefficient indicating the parity of the transform coefficient . Since the pseudo code in the diagram similarly shows the decoding process by turning "code writing" into "decoding", only the tasks mentioned are performed at 16 and 18 respectively. It is contained in one of the first passes 20 in FIG. 18, but can be distributed to two separate passes according to an alternative scheme. Context-adaptive entropy decoding is used to perform both tasks 16 and 8. The context-adaptive entropy decoding can be context-adaptive binary arithmetic decoding.

In the second pass 22 of one or more of the scan transform coefficients, one or more of the greatness flags of the 24 transform coefficients are decoded. Mark, the quantization index of these transform coefficients is not zero. In FIG. 18, there are two such passes 22' and 22". One or more magnitude flag passes 22 need not be separated from one or more first passes 20. In FIG. 18, in pass In 22', gt1_flag decoding 24' is performed, pass 22' is the first pass 20 at the same time, and gt3_flag decoding 24' is performed in a single pass 22". Figure 19 shows that one or more first passes 20 and one or more second passes 22 may coincide. Here, in FIG. 19, decoding 16, 18, and 24 are performed in the same pass. In FIG. 19, there is exemplarily one such pass, which is indicated by the braces indicating the pass as the valid value/parity pass 20 and the magnitude flag pass 22. Similarly, context adaptive entropy decoding is used to perform task 24, and context adaptive entropy decoding can be context adaptive binary arithmetic decoding. In FIG. 18 and FIG. 19, two size level flags are exemplarily decoded, but this is only an example.

In one or more third passes 26 and 27, that is, the two passes separated from each other and passes 20 and 22 under the conditions of Fig. 24 and Fig. 26, the remainder of the quantization index of the transform coefficient is calculated Decoding 28. One or more of the transform coefficients are marked as positive, such as by an indication, that is, the decoding of the coefficients whose size is confirmed/approved, and the sign of the quantization index of the transform coefficient is performed For decoding 30, the quantization index of the transform coefficients is not zero. Use equal-probability entropy decoding to perform tasks 28 and 30, and equal-probability entropy decoding can be equal-probability binary arithmetic decoding. In particular, decoding 28 may involve equal-probability binary arithmetic decoding using binary bits and the remainder of the absolute value of the quantization index of the transform coefficients. One of the transform coefficients or The multiple size flags are just.

The advantages of separating the passes in any of the above-mentioned ways have been promoted above, and it will also become clear from the description of the following possibilities in terms of implementation such as dequantization and/or context derivation, however, these details are regarded as Does not limit the scope of subsequent patent applications. The advantage is to provide a basis for transforming block descriptions that can be efficiently coded using dependent quantization and context-adaptive entropy coding such as context-adaptive binary arithmetic coding.

As shown in Figures 18 and 19, in a first pass 20, for the current scanned transform coefficients, context-adaptive entropy decoding can be used to decode the effective value flags of the current scanned transform coefficients 16, and Then, if the effective value flag indicates that the quantization index of the current scanned transform coefficient is not zero, then as checked at 32, for the current scanned transform coefficient, use context adaptive entropy decoding for the parity flag of the current scanned transform coefficient标 to decode 18. Alternatively, a single first pass 20 may have been used for two flags. In one or more of the second pass 22, that is, 22' and 22" in FIG. 18 and 22 in pass 20 at the same time under the conditions of FIG. 19, if the current scanned transform coefficient is The quantization index is not zero. If this situation is checked at 32, for the current scanned transform coefficients, context adaptive entropy decoding is used to decode the magnitude flags of the current scanned transform coefficients 24', 24". It should be noted that this first size level flag may have different definitions as explained above, and may be gt1_flag or gt3_flag. The decoding 18 under the conditions of Fig. 19 and the decoding 24" under the conditions of Fig. 18 and Fig. 19 are even performed only under the condition that the magnitude flag gt1_flag of the current scanned transform coefficient is already positive. 34 inspections. That is, for the current scanned coefficient, the coding/decoding order of the flags is: the effective value flag, the size level flag gt1_flag, the parity flag, and the size level flag gt2_flag under the condition of FIG. 18, and The size level flag gt3_flag in the situation of FIG. 19.

As mentioned above, from the mentioned flag, for the predetermined transform coefficient, the calculation of the quantization index may involve the sum under the conditions of Fig. 18 and Fig. 19, and the addend is formed by the following: an addend is predetermined The effective value flag of the transform coefficient is formed, an addend is formed by the parity flag of the predetermined transform coefficient, and an addend is formed by the remainder of the predetermined transform coefficient and a magnitude flag (that is, in the situation of FIG. 18 gt2_flag and gt3_flag in the situation of FIG. 19) are formed by twice the sum.

In one or more of the second pass 22, that is, in the conditions of 22" in FIG. 18 and at the same time as pass 20 in the conditions of FIG. 19, the current scanned transform coefficient is decoded, and If the previous size flag of the current scanned transform coefficient is positive, then use context adaptive entropy decoding for another size flag of the current scanned transform coefficient, that is, gt2_flag in the situation of FIG. 18 and that of FIG. 19 Under gt3_flag, decode 24". In the situation of Figure 18, for the predetermined transform coefficient, the absolute value of the quantization index is calculated from the sum. The addend of the sum is formed by the following: one addend is the effective value flag of the predetermined transform coefficient, and one addend is the predetermined For the parity flag of the transform coefficient, an addend is twice the sum of the remainder of the predetermined transform coefficient, the first degree flag and the second degree flag. However, a different definition of the size level flag is used, such as according to Figure 19, for example, but not exclusively, where context adaptation is used Sexual entropy decoding decodes the effective value flag of the currently scanned transform coefficient 16, and then, if the effective value flag indicates that the quantization index of the currently scanned transform coefficient is not zero, as checked at 32, for the current scanned transform coefficient Transform coefficients, use context-adaptive entropy decoding to decode the size flag gt1_flag of the currently scanned transform coefficient 24', and then, if the size flag is positive, as checked at 34, for the current scanned Transform coefficients, use context adaptive entropy decoding to decode the parity flag 18, and for the current scanned transform coefficients, use context adaptive entropy decoding to decode another size flag gt3_flag of the current scanned transform coefficients 24 "For a predetermined transform coefficient, such as the current predetermined transform coefficient, the absolute value of the quantization index is calculated according to the sum. The addend of the sum is formed by the following: an addend is the effective value flag of the predetermined transform coefficient, and an addend is The parity flag of the predetermined transform coefficient, one addend is the size flag gt1_flag of the predetermined transform coefficient, and one addend is twice the sum of the remainder of the predetermined transform coefficient and another size flag gt3_flag. The first size level The decoding of the flags and the second-level flags 24' and 24" can be performed in a separate second pass 22' and 22" as illustrated in FIG. 18 or in one pass as illustrated in FIG. 19 .

As an alternative to Figure 12, it will be possible to use context-adaptive entropy decoding to flag the effective value of the current scanned transform coefficient for the current scanned transform coefficient in one of one or more of the first passes. Perform decoding 16, and then after a first pass, in one or more second passes, use context-adaptive entropy decoding to decode sequences of transform coefficients with more than one magnitude flag 24', 24", and then, After one or more second passes, in the other of the one or more first passes, equal-probability entropy decoding is used to decode the parity flags of the transform coefficients, and the quantization indexes of these transform coefficients are not Is zero.

As mentioned above, coding/decoding can be performed in sub-blocks, so that the transform coefficients are decoded sub-block by sub-block. All passes of the scan position are decoded.

Each coefficient can be reconstructed by the following operation: by selecting one from the "set 0" and "set 1" in Fig. 7 and Fig. 10 in the above example for each transformation coefficient from multiple reconstruction levels. Reconstruct the level set, dequantize the quantization index of each transform coefficient whose quantization index is not zero, and the binarization of each coefficient, that is, all effective values, parity and magnitude flags, remainders and signs have been decoded. According to "setId[state]", this operation is performed by using the current state state. This state in turn depends on the parity of the quantization index of the transform coefficients that precede the individual transform coefficients along the scanning order, that is, for example, by updating the parity of the individual transform coefficients at 40 in FIG. 11 for the previous The state of the dequantized coefficients. After the selection, the respective transform coefficients are dequantized to a level of the selected reconstruction level set, and the level is indexed by setId. This level is the level indexed by the quantization index of each transform coefficient. As described, for each transformation coefficient, the state transition is used to select the reconstruction level set from the reconstruction level set by the following operation: the correct reconstruction level set is selected from multiple reconstruction level sets uniquely based on a state, the state The conversion is for individual transform coefficients, that is, the state is assumed at 42 in Figure 11, and depends on the quantization of the individual transform coefficients The parity of the index updates the state of 40 state transitions for the subsequent transform coefficients in the scan order. Therefore, the scan order is also used here for passes 20, 22, 27, and 30. Examples have been presented above. This is illustrated by arrow 44 in FIG. 4. Similarly, the sub-block subdivision is optional, just like the sub-block decoding of coefficients, that is, each pass can alternatively continue to traverse the sub-blocks before the start of the next pass instead of advancing to the next Execute all passes for a sub-block before execution. There can be four distinct states for state transitions. The conversion can be implemented using the table as illustrated in Table 1 and at 45 where the lookup is performed to generate the states for the following coefficients, or using a trellis diagram as illustrated in Figure 12, for example, where the state appears at 46 .

As described above, multiple reconstruction level sets can be parameterized by means of a predetermined quantization step size Δ as shown in FIGS. 7 and 10. Information about the predetermined quantization step size can be communicated in the data stream. Each of the reconstruction level sets may be for multiple reconstruction level sets using a common horizontal axis t as illustrated in FIGS. 7 and 10 or the predetermined quantization step size that is equal among the multiple reconstruction level sets The multiples of the composition. The number of reconstruction level sets can be two, as illustrated in FIGS. 7 and 10, and the first reconstruction level set may include zero and an even multiple of the predetermined quantization step, such as set 0 in FIGS. 7 and 10. And the second reconstruction level set may include zero and odd multiples of the predetermined quantization step, as shown in set 1 in FIG. 7 and FIG. 10. The first reconstruction level set can be selected for the state values 0 and 1, and the second reconstruction level set can be selected for the state values 2 and 3, as illustrated in Table 1.

It should be noted that if the above dependent quantization scheme is used, after the parity of a certain coefficient is decoded, the quantization set of the next coefficient is defined The state of the state variable is defined or can be determined, and this state can be used to write the effective value and parity flag of the next coefficient.

Now focus on the context selection for performing decoding 16, 18, and 24, and describe the concept of favorable context modeling.

In a preferred embodiment, the adaptive probability model used for conventional code-writing bits is selected from a set of multiple adaptive probability models. The probability model is also called context, and the choice of the probability model is also called context modeling. In a preferred embodiment, the selected probability model depends on one or more of the following properties:

‧Color plane. Generally, the brightness transformation coefficient and the chrominance transformation coefficient have different statistical properties, and therefore, if different probability model sets are used for the brightness and chrominance, the coding efficiency can generally be improved. It is also possible to use a separate set of probability models for each color plane (for example, Y, Cb, Cr).

‧The diagonal position given by the sum of the x and y coordinates inside the transformation block, diag=x+y. On average, the absolute value of the transform coefficient increases as the diagonal position diag decreases. For that reason, if the diagonal position is divided into two or more categories and a separate set of probability models is used for each category, the compression efficiency can generally be improved.

‧Applicable to the state variables of the current transformation coefficients. As mentioned above, the two quantizers Q0 and Q1 have different sets of allowable reconstruction levels. Therefore, the probability mass functions of the two quantizers are significantly different. This aspect has the greatest impact on the probability of sig_flag (which indicates whether the transform coefficient level is not equal to 0). Therefore, if the two quantizers use different probability model sets, the compression efficiency can be improved. As an extension of this concept, state Different values of variables (which can take 4 possible values: 0, 1, 2, 3) use different probability model sets. Or, different sets of probability models can be used for different values of defined functions of state variables. It should be noted that different sets of probability models are used for two different quantizers to represent the special situation of the latter method.

The dependence of state variables may not be suitable for all bits, because it increases the total number of probability models and therefore reduces the speed of probability adaptation. Therefore, the set of different probability models with different values of the function of the state variable can only be used for a subset of the conventional code-writing bits. For example, only for sig_flag's. Or only for sig_flag's and par_flag's (or any other subset).

‧A measure of the activity inside the local neighborhood around the current scan position. Generally, the probability that the absolute value of the current transform coefficient exceeds a certain threshold increases with the activity within the local neighborhood, where the activity refers to, for example, the sum of the absolute transform coefficient levels in the neighborhood. Aspects can be used to improve compression efficiency by using different probability metrics for different local activity metrics. However, it should be noted that since the absolute value is coded in multiple passes, only those data available in a certain pass can be used.

In the following, an exemplary setting of context modeling (choice of probability model) is described in more detail. The example refers to the coding order of the bits specified in Figure 18. However, this concept is not limited to this specific example, and can be easily transferred to the modification of the coding order described above, such as the coding order indicated in FIG. 19, for example.

In order to derive the local activity metric, the local template shown in FIG. 16 can be used in a preferred embodiment. It is also possible to use different templates, such as templates that include more or fewer adjacent scan positions. Usually, better The template used indeed only includes the scanning position that precedes the current scanning position in the coding sequence 44.

Let T(k) denote the set of scanning positions in the local template 52. Then let sumAbs be the sum of absolute values in the local template, which is given by

In addition, let numSig be the absolute number of levels greater than zero in the local template.

It is inferred that the values of absLevel[i] and sig_flag[i] of the position outside the current transformation block are equal to 0.

The metric numSig may have been derived based on the value of sig_flag's. However, the sum of absolute level sumAbs is only available in the third (and fourth) pass 26 and 27. In the first pass 20 and the second pass 22", only a subset of information is coded, and therefore, only this subset of information can be used for context modeling.

According to the preferred embodiment shown in the pseudo code of FIG. 18, the bits sig_flag, par_flag, and gt1_flag are transmitted in the first pass 20/22'. Based on this data of a certain transform coefficient level, the following conclusions can be drawn: ‧If sig_flag is equal to 0, then the transform coefficient level is equal to 0: level=0; ‧If sig_flag is equal to 1, par_flag is equal to 0, and gt1_flag Equal to 0, the transform coefficient level is equal to 1: level=1; ‧If sig_flag is equal to 1, par_flag is equal to 1, and gt1_flag is equal to 0, then the transform coefficient level is equal to 2: level=2; ‧If sig_flag is equal to 1, par_flag is equal to 0, And gt1_flag is equal to 1, then the transform coefficient level is greater than or equal to 3: level>=3; ‧If sig_flag is equal to 1, par_flag is equal to 1, and gt1_flag is equal to 1, then the transform coefficient level is greater than or equal to 4: level>=4; therefore, The minimum value of the absolute transform coefficient level minAbs1[k]=sig_flag[k]+par_flag[k]+2*gt1_flag[k] can be derived at least according to the following formula.

In the second pass, the value of gt2_flag is also known and the following minimum value can be derived accordingly: minAbs2[k]=sig_flag[k]+par_flag[k]+2*(gt1_flag[k]+gt2_flag[k]).

The corresponding sum of the minimum absolute value is represented by sumAbs1 and sumAbs2 and is given by the following formula

and

It is inferred that the values of minAbs1[i] and minAbs2[i] at the position outside the current transformation block are equal to 0.

In this context, it should be noted that the exact equations used to calculate minAbs1[k] and minAbs2[k] (and therefore, The final value of sumAbs1[k] and sumAbs2[k]) depends on which bits are included in the first pass and are included in the semantics (ie, meaning) of these bits. For example, if gt2_flag[k] is included in the first pass or if the meaning of gt1_flag[k] is modified (two alternatives have been described above), the equation used to calculate minAbs1[k] must be correspondingly Change. For example, in the situation of FIG. 19, during the first pass 20/22, the minimum value is minAbs1=sig_flag+gt1_flag+par_flag+2*(gt3_flag+remainder). In any case, minAbs1[k] represents the minimum value of the absolute level that can be derived based on the bits of the code written in the first pass.

The values sumAbs, numSig, sumAbs1 and sumAbs2 or functions of these values can be used as local activity measures for context modeling. A detailed example is described below.

In an alternative embodiment, the values sumAbs, minAbs1, and minAbs2 are determined not based on the transmitted transform coefficient level, but based on the associated multiplication factor (absolute value) of the quantization step. Such data can be derived based on the transformation coefficient levels and associated state variables. Given the absolute level absLevel[k] and the associated state variable state[k], the multiplication factor qIdx[k]=2 * absLevel[k]-( state[k]>>1)

Replacing similar values of minAbs1 and minAbs2 can be derived according to the following formula minQIdx1[k]=2 * minAbs1[k]-(state[k]>>1)

minQIdx2[k]=2 * minAbs2[k]-(state[k]>>1)

Given these equivalent values, the substitute values sumQAbs, sumQAbs1 and sumQAbs2 in place of sumAbs, sumAbs1 and sumAbs2 can be derived according to the following formula:

In the following description, the values sumAbs, sumAbs1, and sumAbs2 are used. But it should be remembered that these equivalent values can be replaced with the values sumQAbs, sumQAbs1 and sumQAbs2. When the values sumQAbs, sumQAbs1, and sumQAbs2 are used, different functions of these equivalent values can be preferably used in the context derivation.

Used for the context modeling of the effective value flag sig_flag

The adaptive probability model used to write the current sig_flag is selected from the probability model set. For simplicity, assume that the available probability model is organized in the following 4-dimensional array probModelSig[cSig][sSig][dSig][aSig], where cSig designation depends on the index of the current color channel, and sSig designation depends on the index of the state variable. The dSig designation depends on the index of the diagonal position (or more generally, the x and y positions) inside the transform block, and the aSig designation depends on the index of the local activity metric. The actual organization of the probability model is the state of actual implementation. It can be organized as a 1-d array, for example, in this case, the combined index can be derived based on the values of cSig, sSig, dSig, and aSig.

In the following, example methods for deriving indexes cSig, sSig, dSig, and aSig are described. However, it should be noted that different ways of deriving these indexes (or parts of indexes) are possible.

Color channel index cSig

In a preferred embodiment, the color channel index cSig is set to be equal to 0 if and only if the current color channel represents the lightness channel (or, more generally, the first color channel). And if and only if the current color channel represents the chroma channel (or, more generally, not the first color channel), cSig is set equal to 1: cSig=(current channel is lumaa? 0:1)

As an alternative, cSig can be set to be equal to 0 for the brightness channel, equal to 1 for the Cb channel, and equal to 2 for the Cr channel.

State index sSig

In the preferred embodiment, the index sSig is set equal to

That means, a set of probability models is used for state variables equal to 0 and 1, the second set is used for state variables equal to 2, and the third set is used for state variables equal to 3.

As an alternative, the index sSig can be set equal to the state variable (sSig=state), in which case a separate probability model set will be used for each possible value of the state variable. Or as another alternative, the index sSig can be set according to sSig=state>>1. In this case, a separate context model set will be used for each of the two quantizers Q0 and Q1 (note that When the state is equal to 0 or 1, the quantizer Q0 is used, and when the state is equal to 2 or 3, the quantizer Q1) is used.

Location index dSig

In the preferred embodiment, the index dSig is set as follows:

‧If the index cSig is equal to 0 (that is, the current color channel represents the brightness channel), set dSig according to the following formula

‧If the index cSig is equal to 1 (that is, the current color channel represents the chroma channel), set dSig according to the following formula

Here, diag represents the diagonal position given by diag=x+y, where x and y represent the x and y coordinates of the current scan position inside the transformation block.

Alternatively, any other clustering of diagonal positions is possible. Alternatively, the (x, y) position inside the transformation block can be divided into multiple categories, and the index dSig can be set equal to the corresponding category index.

Local activity index aSig

Finally, in a preferred embodiment of the present invention, the index aSig aSig=min(5,sumAbs1) is set according to the following formula, where sumAbs1 refers to the minimum absolute transformation given by the data transmitted in the first pass in the partial template The sum of the coefficient levels (see above).

As an alternative, different maximum values or different functions of sumAbs1 can be used.

Used for context modeling of par_flag

The adaptive probability model used to write the current par_flag is selected from the set of probability models. Similarly, for the effective value flag, it is assumed that the available probability model is organized in the following 4-dimensional array probModelPar[cPar][sPar][dPar][aPar], where the cPar specification depends on the index of the current color channel, and the sPar specification depends on the state The index of the variable, dPar designation depends on the index of the diagonal position (or more generally, the x and y position) inside the transformation block, and the index of aPar designation depends on the local activity metric.

Color channel index cPar

Similarly, for the effective value flag, in a preferred embodiment, the color channel index cSig is set according to the following formula: cPar=(current channel is luma? 0:1)

Alternatively, the alternatives described above for cSig can be used.

State index sPar

In a preferred embodiment, the index sPar is set equal to zero. This means that the selected probability model does not depend on state variables.

Alternatively, any of the methods described above for sPar can be used.

Location index dPar

In a preferred embodiment, the index dPar is set as follows:

‧If the index cPar is equal to 0 (that is, the current color channel represents the brightness channel), set dPar according to the following formula

‧If the index cPar is equal to 1 (that is, the current color channel represents the chroma channel), set dPar according to the following formula

Here, diag represents the diagonal position given by diag=x+y, where x and y represent the x and y coordinates of the current scan position inside the transformation block. The Boolean variable firstNonZero specifies whether the current scan position represents the scan position of the first non-zero level in the coding sequence (that is, identified by the x and y coordinates (or similar methods) transmitted after the code block flag Location). Therefore, for the first non-zero level in the coding order, a set of different probability models (independent of the diagonal position) is used.

Alternatively, any other clustering of diagonal positions is possible. Alternatively, the (x, y) position inside the transformation block can be divided into multiple categories, and the index dPar can be set equal to the corresponding category index.

Local activity index aPar

Finally, in a preferred embodiment of the present invention, the index aPar aPar=min(4,sumAbs1-numSig) is set according to the following formula, where sumAbs1 refers to the partial template transmitted in the first pass The sum of the minimum absolute transform coefficient levels given by the data (see above). And numSig refers to the number of non-zero levels in the local template (that is, the number of sig_flag's equal to 1) (see the description above). As an alternative, different maximum values or different functions of sumAbs1 and numSig can be used.

Context modeling of flag gt1_flag

The adaptive probability model used to write the current gt1_flag is selected from the probability model set. Similarly, for the effective value flag, it is assumed that the available probability model is organized in the following 4-dimensional array probModelGt1[cGt1][sGt1][dGt1][aGt1], where the cGt1 specification depends on the index of the current color channel, and the sGt1 specification depends on the state The index of the variable, dGt1 specifies the index that depends on the diagonal position (or more generally, the x and y positions) inside the transform block, and the aGt1 specifies the index that depends on the local activity metric.

In a preferred embodiment of the present invention, the indexes cGt1, sGt1, dGt1, and aGt1 are derived in the same manner as the indexes cPar, sPar, dPar, and aPar described above: cGt1=cPar

sGt1=sPar

dGt1=dPar

aGt1=aPar

It should be noted that using the same context index is different from using the same probability model. Although the derivation of the context model index is the same for gt1_flag and par_flag, the sets of probability models do not intersect. That means, use a set of probability models for par_flag, and for gt1_flag Use another set of probability models.

It should be noted that the context selection of gt1_flag does not depend on the value of par_flag directly in front.

Alternatively, the indexes cGt1, sGt1, dGt1, and aGt1 can be derived by any of the methods described above with respect to the effective value flag.

In addition, the selected probability model may additionally depend on the value of the par_flag directly in front, so that a different set of probability models is used for each of the two parity values. However, this will introduce a direct dependency between par_flag and gt1_flag.

Context modeling of flag gt2_flag

The adaptive probability model used to write the current gt2_flag is selected from the probability model set. Similarly, for the effective value flag, it is assumed that the available probability model is organized in the following 4-dimensional array probModelGt2[cGt2][sGt2][dGt2][aGt2], where the cGt2 specification depends on the index of the current color channel, and the sGt2 specification depends on the state The index of the variable, dGt2 specifies the index that depends on the diagonal position (or more generally, the x and y positions) inside the transform block, and the aGt2 specifies the index that depends on the local activity metric.

In a preferred embodiment, the indexes cGt2, sGt2, dGt2, and aGt2 are derived in the same way as the indexes cGt1, sGt1, dGt1, and aGt1 described above: cGt2=cGt1

sGr2=sGt1

dGt2=dGt1

aGt2=aGt1

Similarly, for gt1_flag, it should be noted that (even if the context index is the same for writing codes gt2_flag and gt2_flag), generally speaking, different probability model sets are used for gt1_flag and gt2_flag. However, as a special situation, it is also possible to use exactly the same probability model for the two flags.

Alternatively, the indexes cGt2, sGt2, dGt2, and aGt2 can be derived by any of the methods described above with respect to the aforementioned flags.

In addition, in order to derive the local activity index aGt2, additional information of gt2_flag in the local neighborhood can be used (in this case, gt2_flag is written in the second pass). For example, the activity index can be set according to the following formula: aGt2=min(4,sumAbs2-numSig), where sumAbs2 refers to the smallest value in the local template given by the data transmitted in the first pass and the second pass The sum of absolute transform coefficient levels (see above).

As an alternative, different maximum values or different functions of sumAbs2, sumAbs1, and numSig can be used.

Turning now to the discussion of FIGS. 18 and 19, in other words, this means the following.

As described, the decoding 16 of the effective value flag of the predetermined transform coefficient may involve selecting the context for performing the decoding 16 depending on the coefficient position of the predetermined transform coefficient within the transform block. This position is shown in Figure 18 and The parameter k in Figure 19 is indexed. Additionally or alternatively, as explained above with respect to FIG. 16, the effective value flag of the predetermined transform coefficient 50 is decoded by the following operation: The set of decoded flags before the effective value flag of the predetermined transform coefficient of a set of adjacent transform coefficients 51 determines the local activity, and depends on the local activity selection context. As shown in Figure 18 and Figure 19, for example, one of the effective value flag, the parity flag, and one or more size flags, or all the size flags under the condition of Figure 19 can be in one pass Secondly, code/decode so that the set of flags decoded for the coefficient 51 in the template 52 includes such flags as decoded for the set of adjacent transform coefficients 51 in the template 52, and the activity can be based on the adjacent transform Each of the coefficients 51 is calculated with the sum of the addends. The addends indicate the minimally assumed index or the minimally assumed reconstruction level for each adjacent transformation coefficient 51. The index or reconstruction level system The decision is based on the previously decoded flags for the respective adjacent transform coefficients. The minimally assumed value means the minimum threshold value for the quantization index or reconstruction level. The respective adjacent coefficients are minimally assumed based on the analysis of the previously derived flags for the respective adjacent coefficients 51 Index or restructure hierarchy. The minimum value can be defined in an absolute sense. In the situation of FIG. 19, for example, this minimum value can be calculated as sig_flag+gt1_flag+par_flag+2*gt3_flag of each coefficient 51, because only the remainder can be missing any coefficient. For the previously written code/decoding flag set because gt1_flag is zero, the coefficient 51 does not include certain flags such as par_flag and gt3_flag, and each flag has a default value of zero. Can use context to code/decode predetermined transform coefficients The effective value flag, the context is even additionally or alternatively selected depending on the state 46 of the state transition assuming the predetermined transformation coefficient. For example, one context set may be defined and selected for states 0 and 1, another context set may be defined and selected for state 2, and a third context set may be defined and selected for state 3. Within the selected context set, the selection of the final use context can then be performed, such as depending on the local activity using any of the previously mentioned dependencies. You can use the currently selected reconstruction set, such as setId, instead of the state.

For parity flags, context selectivity can be similarly designed. That is, a context selected depending on one or more of the following can be used to code/decode the parity flag of the predetermined transform coefficient: 1) the coefficient position of the predetermined transform coefficient, 2) the local activity, which is Determined based on the set of decoded flags before the parity flags of the predetermined transform coefficients of a group of adjacent transform coefficients 51 in the local template 52 surrounding the predetermined transform coefficient 50, 3) within the local template 52 surrounding the predetermined transform coefficient The number of transform coefficients 51, the reconstruction level of these transform coefficients is not zero, and the selection context depends on the number of local activities and/or transform coefficients, or 4) the difference between the number of local activities and the number of transform coefficients.

For the size degree flags such as any one of gt1_flag, gt2_flag, and gt3_flag, context selectivity can be similarly designed. That is, a context selected depending on one or more of the following can be used to code/decode the magnitude flag of the predetermined transform coefficient: 1) the coefficient position of the predetermined transform coefficient, 2) the local activity, which It is based on the local template 52 surrounding or located at the predetermined transform coefficient The magnitude of the predetermined transform coefficients of a group of adjacent transform coefficients 51 is determined by a set of decoded flags before the flag, 3) the number of transform coefficients 51 in the local template 52, and the reconstruction level of these predetermined transform coefficients is not Is zero, and/or 4) The difference between the number of local activities and the number of transform coefficients.

Regarding the binarization of the remainder, the following advantageous concepts can be used.

In a preferred embodiment, the remainder of the syntax element is coded in the bypass mode of the arithmetic coding engine (see above). The compression efficiency depends on the binarization used. In HEVC, a type of binary code called Columbus-Rice code is used to code similar syntax elements. The codes of this category are parameterized by so-called Rice parameters. In HEVC, the Rice parameter is adjusted during coding so that the binarization used depends on the previous syntax elements.

In a preferred embodiment, Golomb-Rice codes of the same (or very similar) code type to HEVC are used to binarize the rest of the syntax elements. Compared with HEVC, after dividing the level information of the remainder by two, the derivation of Rice parameters must be modified.

In a preferred configuration, the sumAbs of the absolute value of the adjacent levels covered by the partial template is used to derive the Rice parameter (see above). In a particularly preferred embodiment, the Rice parameter RP is derived according to the following formula.

In other configurations, the threshold value used to switch Rice parameters can be modified. Or it can be derived based on other activity measures of the local template. number. In addition, it may be additionally specified that the Rice parameter is not allowed to become smaller in the sub-block.

As an alternative, the Rice parameter can be modified in a similar way as in HEVC.

That is, the Rice parameter selection can be performed when the 28 remainder is decoded.

In the embodiment of FIG. 18 described above, for example, the parity syntax element is transmitted directly after the effective value information. In this configuration, the direct inter-bit dependency only exists between the effective value information and the parity information. After the coding/decoding of the parity information, the gt1_flag is coded, thereby stopping the coding of the level information of the current scanning position in the first pass. After finishing the first scan pass 20/22', gt2_flags is transmitted in the second scan pass 22'. Finally, the remainder is transmitted in scan pass 26. This design minimizes the direct bit-subdependency with valid values and parsing syntax elements. In addition, more levels of information can be modeled and evaluated against the context of the effective value flag, thereby achieving higher compression efficiency. The context modeling of the parity flag is exactly the same as the context modeling of the gt1 and gt2 flags, resulting in less logic and therefore less complexity.

In an alternative embodiment, the parity information is transmitted first before the absolute level information, as illustrated in the pseudo code of FIG. 20. Here, the effective value flag is transmitted in the pass 20', then the parity flag is transmitted in the pass 20", and then the first magnitude flag is transmitted in the pass 22', and then the flag is transmitted in the pass 22" The second largest degree flag. When the parity information is equal to 0, only the effective value flag is required, and when the parity flag is equal to 1, the effective value flag is inferred, etc. At 1. The advantage of this configuration is that there is no direct inter-bit dependency. After the parity flag is transmitted in the first pass, the values of the state variables and therefore the quantizers used for the individual transform coefficients are known.

The number of passes between the pass using sig_flags and the pass using the remainder's pass-by-code writing can be selected to be equal to zero or equal to any value greater than zero under the condition of FIG. 19. In addition, multiple flags can be transmitted in one pass.

The further modifications described above are related to the following. The selected probability model used for par_flag can depend on one or more of the following parameters: ‧color channel; ‧state variables; ‧parity flag in the local template surrounding the current scan position; ‧scan position (e.g. , By the diagonal position of the cluster).

The context modeling for grammatical elements that follow parity can be adjusted based on parity information, so that different sets of context models are used depending on parity. In addition, the syntax elements may depend on any of the parameters described before the parity flag.

In this configuration, the number of bits using the context model is increased, for example, to six (the last context-coded syntax element is then GT5). Then, the parity information is transmitted in the bypass mode, which is almost equal due to the conditional probability. This configuration has the following benefits: it can achieve the compression efficiency of a design that transmits complete information for each scanning position.

Regarding entropy in the case of non-dependent scalar quantification Write the code and state the following.

Even though the entropy coding described above is particularly advantageous for coding with dependent quantization, it can also be advantageously applied to coding with conventional non-dependent quantization. The only aspect that needs to be modified is the derivation of state variables. In fact, for conventional non-dependent quantization, the state variable can always be set equal to zero. It should be noted that the quantizer Q0 represents a conventional uniform reconstruction quantizer.

Back to the conventional scalar quantization can also be achieved by modifying the state transition table to the following formula state_trans_table[4][2]={{0,0},{0,0},{0,0},{0,0 }}

Since the state transition table can be expressed as a single 16-bit integer (see the description above), the same implementation can be used for dependent and non-dependent scalar quantization. The quantization method used can be configured by representing the 16-bit value of the state transition table. Therefore, the described method for entropy coding is also suitable for codecs that support switching between dependent quantization and non-dependent quantization (for example, at the sequence, image, tile, image block, or block level).

The other embodiments relate to the following.

1. A device for decoding a block of transform coefficients, which is configured with: a) In the first pass of scanning one or more of the transform coefficients, context-adaptive binary arithmetic is used Decoding decodes one of the effective value flags of the transform coefficients, one of the parity flags of the transform coefficients, and one or more magnitude flags of the transform coefficients whose quantization index is not zero. The effective value flag indicates the Whether the quantization index of one of the transform coefficients is zero, the parity flag indicates that one of the transform coefficients is in the same position; and b) In one or more second passes, using equal-probability binary arithmetic decoding on the transform coefficients, one of the absolute value of the quantization index, one of the remainder, and one of the binary bits of the transform coefficients A symbol of the quantization index is decoded, the one or more magnitude flags of the transform coefficients are positive, and the quantization index of the transform coefficients is not zero.

2. The device as in embodiment 1, which is configured to use context-adaptive binary arithmetic decoding for a current scanned transform coefficient in a first pass to decode the effective value flag of the current scanned transform coefficient If the effective value flag indicates that the quantization index of the current scanned transform coefficient is not zero, then for the current scanned transform coefficient, use context-adaptive binary arithmetic decoding for the current scanned transform coefficient The parity flag is decoded.

3. The device of embodiment 1 or 2, which is configured to target a current scanned transform coefficient in the one or more first passes, if the quantization index of the current scanned transform coefficient is not zero , Then use context adaptive binary arithmetic decoding to decode the first magnitude flag of one of the current scanned transform coefficients.

4. As the device of embodiment 3, it is configured to calculate the absolute value of the quantization index based on a sum for a predetermined transform coefficient, and the addend of the sum is formed by: the effective value of the predetermined transform coefficient The flag, the parity flag of the predetermined transform coefficient, and twice the sum of the remainder of the predetermined transform coefficient and one of the first magnitude flags.

5. As the device of embodiment 3, it is configured to, in the one or more second passes, for a current scanned transform coefficient, if the current If the first magnitude flag of the scanned transform coefficient is positive, the second magnitude flag of one of the current scanned transform coefficients is decoded using context adaptive binary arithmetic decoding.

6. The device of embodiment 5 is configured to calculate the absolute value of the quantization index based on a sum for a predetermined transform coefficient, and the addend of the sum is formed by: the effective value of the predetermined transform coefficient The value flag, the parity flag of the predetermined transform coefficient, the first degree flag of the predetermined transform coefficient, and the remainder of the predetermined transform coefficient twice the sum of the second degree flag.

7. The device of any one of embodiments 1 to 6, which is configured to perform each of the one or more second passes after each of the one or more first passes.

8. The device as in any one of the preceding embodiments, in which the transform coefficients of a transform block are divided into sub-blocks, and the transform coefficients are decoded sub-block by sub-block, wherein all the scanning positions of a sub-block are The pass is decoded before the first pass of the next sub-block is decoded.

9. The device as in any one of embodiments 1 to 8, which is configured to dequantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depending on a scan order before each For the parity of the quantization index of the transform coefficient of the individual transform coefficient, select a reconstruction level set from a plurality of reconstruction level sets for the individual transform coefficient, and dequantize the individual transform coefficient to a level of the reconstruction level set Above, the level is indexed by the quantization index of the respective transform coefficient.

10. The device as in Example 9, which is configured for the The individual transform coefficients are selected from a plurality of reconstruction level sets using a state transition by the following operation: the state transition is uniquely based on the assumption of a state from the plurality of reconstruction level sets for the individual transform coefficients. The reconstruction level set selects the reconstruction level set, and depends on the parity of the quantization index of the respective transform coefficient, and updates the state of the state transition for a transform coefficient subsequent to the scan order.

11. The device of embodiment 10, which is configured to perform the one or more first passes and/or the one or more second passes along the scanning order.

12. The device of embodiment 10 or 11 is configured to perform state transitions between four different states.

13. The device of any one of embodiments 9 to 12, which is configured to parameterize and stream the multiple (50) reconstruction level sets (52) by means of a predetermined quantization step size ( 14) Derive information about the predetermined quantization step.

14. The device of any one of embodiments 9 to 13, wherein each of the multiple (50) reconstruction level sets (52) is composed of an integer multiple of a predetermined quantization step size, and the predetermined quantization step size The set of reconstruction levels (52) is constant for these multiple (50) reconstruction levels.

15. The device of any one of embodiments 9 to 14, wherein the number of reconstruction level sets (52) in the plurality (50) of reconstruction level sets (52) is two, and the multiple The structure level set includes: a first reconstruction level set including zero and an even multiple of a predetermined quantization step, and a second reconstruction level set including zero and an odd multiple of the predetermined quantization step.

16. The device of embodiment 15, which is configured to select the first reconstruction level set for state values 0 and 1, and select the second reconstruction level set for state values 2 and 3.

17. The device of any one of embodiments 1 to 16, which is configured to use context-adaptive binary arithmetic by selecting a context depending on the position of a coefficient of the predetermined transform coefficient inside the transform block Decoding decodes the effective value flag of a predetermined transform coefficient.

18. The device as in any one of embodiments 1 to 17, which is configured to use context adaptive binary arithmetic decoding to decode the effective value flag of a predetermined transform coefficient by the following operation: A set of decoded flags before the effective value flag of a predetermined transform coefficient of a group of adjacent transform coefficients in a local template of the predetermined transform coefficient determines a local activity, and a context is selected depending on the local activity.

19. The device of embodiment 18 is configured to decode the effective value flag, the parity flag, and the one or more magnitude flags in a first pass, so that the flag set Contains the effective value flag, the parity flag, and the one or more magnitude flags decoded for the set of adjacent transform coefficients, and is based on a sum of one of the addends of each of the adjacent transform coefficients To calculate the local activity, the addends indicate a minimally assumed index or a minimally assumed reconstruction level for one of the adjacent transformation coefficients, and the index or reconstruction level is based on the corresponding Determine the effective value flag, the parity flag, and the one or more magnitude flags of the adjacent transform coefficient decoding.

20. As the device of any one of Examples 1 to 19, it has been Combine with: a) For the individual transform coefficient, select the reconstruction level set from a plurality of reconstruction level sets using a state transition by performing the following operations: uniquely assume for the individual transform coefficient based on the state transition A state selects the reconstruction level set from the plurality of reconstruction level sets, and depends on the parity of the quantization index of the respective transform coefficient, and updates the state transition for a subsequent transform coefficient in the scanning order The state; b) perform the one or more first passes along the scan order; and c) use context adaptation by selecting a context depending on the state transition for the state assumed by the predetermined transform coefficient (13') The binary arithmetic decoding decodes the effective value flag of a predetermined transform coefficient.

21. The device of embodiment 20, which is configured to decode the effective value flag of a predetermined transform coefficient by using context adaptive entropy decoding by selecting a context set including the context depending on the state, and The context is selected from the set of contexts depending on the local activity around one of the predetermined transform coefficients or the position of one of the predetermined transform coefficients.

22. The device of embodiment 21, wherein a first context set is selected for states 0 and 1, a second context set is selected for state 2, and a third context set is selected for state 3.

23. The device as in any one of embodiments 1 to 22, which is configured with: a) Dequantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depending on the first scan order In the parity of the quantization index of the transform coefficient of the respective transform coefficient, a reconstruction level set is selected from a plurality of reconstruction level sets for the respective transform coefficient, and the respective transform coefficient is dequantized to the reconstruction level set On one level, the level consists of The quantization indexing of the individual transform coefficients; b) perform the one or more first passes along the scanning order; and c) select one by depending on the set of reconstruction levels selected for a predetermined transform coefficient Context-adaptive binary arithmetic decoding is used to decode the effective value flag of the predetermined transform coefficient.

24. The device of any one of embodiments 1 to 23, which is configured to use context-adaptive binary arithmetic by selecting a context depending on the position of a coefficient of the predetermined transform coefficient inside the transform block Decoding decodes the parity flag of a predetermined transform coefficient.

25. The device as in any one of embodiments 1 to 24, which is configured to use context-adaptive binary arithmetic decoding to decode the parity flag of a predetermined transform coefficient by the following operation: A set of flags decoded before the parity flag of a predetermined transform coefficient of a set of adjacent transform coefficients in a local template of a predetermined transform coefficient determines a local activity, and/or determines a local activity around the predetermined transform coefficient in the local template For a number of transform coefficients, the reconstruction level of the predetermined transform coefficients is not zero; and a context is selected depending on the local activity and/or the number of transform coefficients.

26. The device of embodiment 25, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients.

27. The device of embodiment 25 or 26, which is configured to decode the effective value flag, the parity flag, and the one or more magnitude flags in a first pass, so that the flag The target set contains the Set the effective value flag, the parity flag, and the one or more magnitude flags decoded by adjacent transform coefficients, and calculate the sum based on a summation of one of the adjacent transform coefficients Activity, the addends indicate a minimally assumed index or a minimally assumed reconstruction level for one of the adjacent transform coefficients, the index or reconstruction level based on decoding for the adjacent transform coefficients The effective value flag, the parity flag, and the one or more magnitude flags are determined.

28. The device of any one of embodiments 1 to 27, which is configured to use context-adaptive binary arithmetic by selecting a context depending on the position of a coefficient of the predetermined transform coefficient inside the transform block Decoding decodes a first size flag of one of one or more size flags of a predetermined transform coefficient.

29. The device as in any one of embodiments 1 to 28, which is configured to use context-adaptive binary arithmetic decoding to decode the first magnitude flag of a predetermined transform coefficient by the following operations: based on A set of decoded flags before the first magnitude flag of a predetermined transform coefficient of a group of adjacent transform coefficients in a local template surrounding the predetermined transform coefficient determines a local activity, and/or determines a local activity around the predetermined transform A number of transform coefficients in the local template of coefficients, the reconstruction level of the predetermined transform coefficients is not zero; and a context is selected depending on the local activity and/or the number of transform coefficients.

30. The device of embodiment 29, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients inside the transform block.

31. The device of embodiment 29 or 30 is configured to decode the effective value flag, the parity flag, and the one or more magnitude flags in a first pass, so that the flag The flag set includes the effective value flag, the parity flag, and the one or more magnitude flags decoded for the set of adjacent transform coefficients, and is based on the addend of one of the adjacent transform coefficients The activity is calculated as a sum, and the addends indicate a minimally assumed index or a minimally assumed reconstruction level for one of the adjacent transform coefficients, the index or reconstruction level being based on the The effective value flag, the parity flag, and the one or more magnitude flags of adjacent transform coefficients are decoded.

32. The device as in any one of embodiments 1 to 31, which is configured to use a Columbus-Rice code and use a Rice parameter to decode the remainder of a predetermined transform coefficient, the Rice parameter depends on the surrounding A sum of the addends of each of a group of adjacent transform coefficients in a local template of the predetermined transform coefficient, and the addend depends on a quantization index or a reconstruction level of the respective adjacent transform coefficients.

33. A device for encoding a block of transform coefficients, which is configured to: a) use context adaptive binary arithmetic in scanning one or more first passes of the transform coefficients Encoding one or more of the effective value flag of the current transform coefficient (for example, sig_flag[k]), the parity flag of one of the transform coefficients (for example, par_flag[k]), and the transform coefficient whose quantization index is not zero A size level flag (for example, gt#_flag[k]) is coded, and the effective value flag indicates whether one of the quantization indexes of the current transform coefficient (for example, level[k]) is zero, and the parity flag indicates Show that one of the transform coefficients is in the same position; and b) in one or more third passes, use equal-probability binary arithmetic coding for the remainder of the quantization index of the transform coefficients (for example, remainder[k]) And one of the quantization indexes of the transform coefficients (for example, sign[k]) is coded, the one or more magnitude flags of the transform coefficients are positive, and the quantization index of the transform coefficients is not zero.

34. The device of embodiment 33, which is configured to, in a first pass, for a current scanned transform coefficient, a) Use context adaptive binary arithmetic coding to be valid for the current scanned transform coefficient Value flag for decoding, and b) if the effective value flag indicates that the quantization index of the current scanned transform coefficient is not zero, then for the current scanned transform coefficient, use context-adaptive binary arithmetic coding for the current The parity flag of the scan transform coefficient is coded.

35. The device of embodiment 33 or 34, which is configured to target a current scanned transform coefficient in the one or more second passes, if the quantization index of the current scanned transform coefficient is not zero , Then use context-adaptive binary arithmetic coding to encode the first magnitude flag of one of the current scanned transform coefficients.

36. The device of embodiment 35, which is configured so that for a predetermined transform coefficient, the quantization index can be obtained according to a sum, and the addend of the sum is formed by: the effective value flag of the predetermined transform coefficient The parity flag of the predetermined transform coefficient, and twice the sum of the remainder of the predetermined transform coefficient and one of the first magnitude flags.

37. As the device of embodiment 36, it is assembled in the one In the second or multiple second passes, for a current scanned transform coefficient, if the first magnitude flag of the current scanned transform coefficient is positive, then context adaptive binary arithmetic coding is used for the current scanned transform coefficient One of the transform coefficients is encoded with a second magnitude flag.

38. The device of embodiment 37, which is configured for a predetermined transform coefficient to obtain the quantization index according to a sum, and the addend of the sum is formed by: the effective value flag of the predetermined transform coefficient , The parity flag of the predetermined transform coefficient, the first degree flag of the predetermined transform coefficient, and twice the sum of the remainder of the predetermined transform coefficient and the second degree flag.

39. The device of any one of embodiments 33 to 38, which is configured to perform each of the one or more second passes after each of the one or more first passes.

40. The device as in any one of the preceding embodiments, wherein the transform coefficients of a transform block are divided into sub-blocks, and the transform coefficients are decoded sub-block by sub-block, wherein all the scan positions of a sub-block are The pass is decoded before the first pass of the next sub-block is decoded.

41. The device of any one of embodiments 33 to 40, which is configured to quantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depending on the order of scanning before the individual For the parity of the quantization index of the transform coefficient of the transform coefficient, a reconstruction level set is selected from a plurality of reconstruction level sets for the respective transform coefficient, and the respective transform coefficient is quantized to a level of the reconstruction level set, The level is indexed by the quantization index of the respective transform coefficient.

42. The device of embodiment 47, which is configured to select the reconstruction level set from a plurality of reconstruction level sets using a state transition for the respective transformation coefficient by the following operation: uniquely based on the state transition For the respective transform coefficient, assume a state to select the reconstruction level set from the plurality of reconstruction level sets, and depend on the parity of the quantization index of the respective transform coefficient, for the latter in the scan order A transformation coefficient updates the state of the state transition.

43. The device of embodiment 42, which is configured to perform the one or more first passes and the one or more second passes along the scanning order.

44. The device of embodiment 42 or 43 is configured to perform state transitions between four different states.

45. The device of any one of embodiments 40 to 43, which is configured to parameterize the multiple reconstruction level sets by means of a predetermined quantization step size and derive information about the predetermined quantization step size from the data stream的信息。 Information.

46. The device of any one of embodiments 40 to 45, wherein each of the plurality of reconstruction level sets is composed of a multiple of a predetermined quantization step, and the predetermined quantization step is for the plurality of reconstructions The set of levels is constant.

47. The device of any one of embodiments 40 to 46, wherein the number of reconstruction level sets in the plurality of reconstruction level sets is two, and the plurality of reconstruction level sets includes: a first layer The structure level set includes zero and an even multiple of a predetermined quantization step, and a second reconstruction level set includes zero and an odd multiple of the predetermined quantization step.

48. The device as in Example 47, which is equipped with a targeted state The state values 0 and 1 select the first reconstruction level set, and for the state values 2 and 3, the second reconstruction level set is selected.

49. The device of any one of embodiments 33 to 48, which is configured to use context-adaptive binary arithmetic by selecting a context depending on the position of a coefficient of the predetermined transform coefficient inside the transform block The encoding encodes the effective value flag of a predetermined transform coefficient.

50. The device as in any one of embodiments 33 to 49, which is configured to use context-adaptive binary arithmetic coding to encode the effective value flag of a predetermined transform coefficient by the following operation: A set of decoded flags before the effective value flag of a predetermined transform coefficient of a group of adjacent transform coefficients in a local template of the predetermined transform coefficient determines a local activity, and a context is selected depending on the local activity.

51. The device of embodiment 50 is configured to encode the effective value flag, the parity flag, and the one or more magnitude flags in a first pass, so that the flag set Contains the effective value flag, the parity flag, and the one or more magnitude flags decoded for the set of adjacent transform coefficients, and is based on a sum of one of the addends of each of the adjacent transform coefficients To calculate the local activity, the addends indicate a minimally assumed index or a minimally assumed reconstruction level for one of the adjacent transformation coefficients, and the index or reconstruction level is based on the corresponding Determine the effective value flag, the parity flag, and the one or more magnitude flags of the adjacent transform coefficient decoding.

52. The device of any one of embodiments 33 to 51, which is configured with: a) For the respective transform coefficient, use the following operations A state transition is performed to select the reconstruction level set from multiple reconstruction level sets: the reconstruction level set is selected from the multiple reconstruction level sets uniquely based on the state transition assuming a state for the respective transformation coefficients, Depending on the parity of the quantization index of the respective transform coefficient, update the state of the state transition for a transform coefficient subsequent to the scan order; b) perform the one or more first passes along the scan order And c) by using context adaptive binary arithmetic coding to encode the effective value flag of a predetermined transform coefficient by selecting a context depending on the state of the state transition assumed for the predetermined transform coefficient (13').

53. The device of embodiment 52, which is configured to use context-adaptive entropy coding to encode the effective value flag of a predetermined transform coefficient by selecting a context set that includes the context depending on the state, and The context is selected from the set of contexts depending on the local activity around one of the predetermined transform coefficients or the position of one of the predetermined transform coefficients.

54. The device of embodiment 53, wherein a first context set is selected for states 0 and 1, a second context set is selected for state 2, and a third context set is selected for state 3.

55. The device of any one of embodiments 33 to 54, which is configured with: a) quantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depends on the quantization index along a scan order For the parity of the quantization index of the transform coefficient of the respective transform coefficient, a reconstruction level set is selected from a plurality of reconstruction level sets for the respective transform coefficient, and the respective transform coefficient is dequantized to the reconstruction level set On a level, the level is indexed by the quantization index of the respective transform coefficient; b) execute along the scan order Perform the one or more first passes; and c) use a context-adaptive binary arithmetic coding pair by selecting a context depending on the reconstruction level set (48) selected for the predetermined transform coefficient (13') The effective value flag of a predetermined transform coefficient is coded.

56. The device of any one of embodiments 33 to 55, which is configured to use context-adaptive binary arithmetic by selecting a context depending on the position of a coefficient of the predetermined transform coefficient inside the transform block The encoding encodes the parity flag of a predetermined transform coefficient.

57. The device as in any one of embodiments 33 to 56, which is configured to use context-adaptive binary arithmetic coding to encode the parity flag of a predetermined transform coefficient by the following operation: A set of flags decoded before the parity flag of a predetermined transform coefficient of a set of adjacent transform coefficients in a local template of a predetermined transform coefficient determines a local activity, and/or determines a local activity around the predetermined transform coefficient in the local template For a number of transform coefficients, the reconstruction level of the predetermined transform coefficients is not zero; and a context is selected depending on the local activity and/or the number of transform coefficients.

58. The device of embodiment 57, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients.

59. The device of embodiment 57 or 58, which is configured to encode the effective value flag, the parity flag, and the one or more magnitude flags in a first pass, so that the flag The flag set includes the effective value flag, the parity flag, and the one decoded for the group of adjacent transform coefficients. Or multiple magnitude flags, and the activity is calculated based on a sum of the addends of each of the adjacent transform coefficients, the addends indicating the minimum for one of the adjacent transform coefficients A hypothetical index or a minimally hypothetical reconstruction level based on the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients Standard and judgment.

60. The device of any one of embodiments 33 to 59, which is configured to use context-adaptive binary arithmetic by selecting a context depending on the position of a coefficient of the predetermined transform coefficient inside the transform block The encoding encodes the first size flag of one of one or more size flags of a predetermined transform coefficient.

61. The device of any one of embodiments 33 to 60, which is configured to use context-adaptive binary arithmetic coding to encode the first magnitude flag of a predetermined transform coefficient by the following operations: based on A set of decoded flags before the first magnitude flag of a predetermined transform coefficient of a group of adjacent transform coefficients in a local template surrounding the predetermined transform coefficient determines a local activity, and/or determines a local activity around the predetermined transform A number of transform coefficients in the local template of coefficients, the reconstruction level of the predetermined transform coefficients is not zero; and a context is selected depending on the local activity and/or the number of transform coefficients.

62. The device of embodiment 61, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients.

63. Such as the equipment of embodiment 61 or 62, which is equipped with In a first pass, the effective value flag, the parity flag, and the one or more magnitude flags are sequentially encoded, so that the flag set includes the decoded set of adjacent transform coefficients The effective value flag, the parity flag, and the one or more magnitude flags are calculated based on a sum of the addends of each of the adjacent transformation coefficients, and the addends are indicated by A minimally assumed index or a minimally assumed reconstruction level at one of the adjacent transform coefficients, the index or reconstruction level being based on the effective value flag, the The parity flag and the one or more size and degree flags are determined.

64. The device of any one of embodiments 33 to 63, which is configured to use a Columbus-Rice code and use a Rice parameter to encode the remainder of a predetermined transform coefficient, the Rice parameter depends on the surrounding A sum of the addends of each of a group of adjacent transform coefficients in a local template of the predetermined transform coefficient, and the addend depends on a quantization index or a reconstruction level of the respective adjacent transform coefficients.

65. A method which is executed by a device as in any one of the above embodiments.

66. A computer program for sending instructions to a computer, and the computer executes the computer program to perform the method in embodiment 65.

67. A data stream generated by the device as in any one of Embodiments 33 to 64.

Although some aspects have been described in the context of the device, it is obvious that these aspects also represent the description of the corresponding method, where the block or device corresponds to the method step or the feature of the method step. Similarly, one of the method steps The aspect described in the context also represents the description of the characteristics of the corresponding block or item or the corresponding device. Some or all of the method steps can be executed by (or using) hardware devices (such as microprocessors, programmable computers, or electronic circuits). In some embodiments, one or more of the most important method steps can be executed by this device.

The encoded data stream of the present invention can be stored on a digital storage medium or can be transmitted on a transmission medium such as a wireless transmission medium or a wired transmission medium such as the Internet.

Depending on certain implementation requirements, the embodiments of the present invention can be implemented in hardware or in software. It can be implemented using digital storage media such as floppy disks, DVDs, Blu-rays, CDs, ROMs, PROM, EPROM, EEPROM or flash memory, on which are stored electronic data that can cooperate with (or can cooperate) with programmable computer systems. Read the control signal to execute each method. Therefore, the digital storage medium can be computer readable.

Some embodiments according to the invention comprise a data carrier with electronically readable control signals that can cooperate with a programmable computer system to perform one of the methods described herein.

Generally speaking, the embodiments of the present invention can be implemented as a computer program product with a program code. When the computer program product runs on a computer, the program code is operatively used to execute one of these methods. The program code can be stored on a machine-readable carrier, for example.

Other embodiments include a computer program stored on a machine-readable carrier for executing one of the methods described herein.

In other words, the embodiment of the method of the present invention is therefore a computer program A formula, which has code for executing one of the methods described in this article when the computer program is executed on the computer.

Therefore, another embodiment of the method of the present invention is a data carrier (or a digital storage medium, or a computer-readable medium), which includes a computer program recorded on it for performing one of the methods described herein. Data carriers, digital storage media, or recorded media are usually tangible and/or non-transitory.

Therefore, another embodiment of the method of the present invention represents a data stream or signal sequence of a computer program used to execute one of the methods described herein. The data stream or signal sequence may for example be configured to be connected via data communication, for example via the Internet.

Another embodiment includes processing components, such as a computer or programmable logic device that is configured or adapted to perform one of the methods described herein.

Another embodiment includes a computer on which a computer program for executing one of the methods described herein is installed.

Another embodiment according to the present invention includes a device or system configured to (for example, electronically or optically) transmit a computer program for performing one of the methods described herein to a receiver. The receiver can be, for example, a computer, a mobile device, a memory device, or the like. The device or system may, for example, include a file server for sending computer programs to the receiver.

In some embodiments, a programmable logic device (eg, a field programmable gate array) can be used to perform some or all of the functionality of the methods described herein. In some embodiments, the field programmable gate array can be combined with The microprocessors cooperate to perform one of the methods described herein. Generally, these methods are preferably executed by any hardware device.

The devices described in this article can be implemented using hardware devices, computers, or a combination of hardware devices and computers.

The device described herein or any component of the device described herein may be implemented at least partially in hardware and/or in software.

The method described in this article can be performed using hardware equipment or using a computer or a combination of hardware equipment and a computer.

The method described herein or any component of the device described herein may be executed at least in part by hardware and/or software.

The above embodiments only illustrate the principle of the present invention. It should be understood that modifications and changes to the configuration and details described herein will be obvious to those skilled in the art. Therefore, it is intended to be limited only by the scope of the following patent applications, and not limited by the specific details presented by the description and explanation of the embodiments herein.

references

[1] ITU-T and ISO|IEC, "Advanced video coding for audiovisual services," ITU-T Roc. H.264 and ISO|IEC 14406-10 (AVC), 2003.

[2] ITU-T and ISO|IEC, "High efficiency video coding," ITU-T Rec. H.265 and ISO|IEC 23008-10 (HEVC), 2013.

[3] Abrecht, et. al., “Description of SDR, HDR, and 360° video coding technology proposal by Fraunhofer HHI,” Joint Video Experts Team (JVET), doc. JVET-J0014, Apr. 2018.

Claims (82)

A device for decoding a block of transform coefficients. The device is configured to use context-adaptive entropy decoding to decode the transform coefficients in one or more first passes of the transform coefficients. A valid value flag is decoded, the valid value flag indicates whether a quantization index of the transform coefficients is zero, and the parity flag of one of the transform coefficients is decoded, and the parity flag indicates the value of the transform coefficients. Together, in one or more second passes of scanning the transform coefficients including or not including one or more of the one or more first passes, the quantization is performed using context-adaptive entropy decoding The transform coefficients whose index is not zero are decoded with one or more magnitude flags. In one or more third passes, the following are decoded using equal-probability entropy decoding. The quantization index of the transform coefficients A remainder, the one or more magnitude flags of the transform coefficients are positive, and a sign of the quantization index of the transform coefficients, the quantization index of the transform coefficients is not zero. For example, the device of claim 1, which is configured to decode the effective value flag of the current scanned transform coefficient using context-adaptive entropy decoding for a current scanned transform coefficient in a first pass , And if the effective value flag indicates the current scanned transform coefficient If the quantization index is not zero, then for the current scanned transform coefficient, use context adaptive entropy decoding to decode the parity flag of the current scanned transform coefficient. Such as the device of claim 1, which is configured to use context-adaptive entropy decoding for a current scanned transform coefficient in one of the one or more first passes The parity flag is decoded, and in the other of the one or more first passes, for a current scanned transform coefficient, if the parity flag indicates the quantization index of the current scanned transform coefficient If it is an even number, for the current scanned transform coefficient, use context adaptive entropy decoding to decode the effective value flag of the current scanned transform coefficient. For example, the device of claim 1, which is configured to, in the one or more second passes, for a current scanned transform coefficient, if the quantization index of the current scanned transform coefficient is not zero, use the context The adaptive entropy decoding decodes the first magnitude flag of one of the currently scanned transform coefficients. For example, the device of claim 4 is configured to calculate the quantization index based on a sum for a predetermined transform coefficient, and the addend of the sum is formed by the following: the effective value flag of the predetermined transform coefficient, The parity flag of the predetermined transform coefficient, the remainder of the predetermined transform coefficient and the sum of one of the first magnitude flags are twice the sum. For example, the device of claim 4, which is configured for a current scanned transform coefficient in the one or more second passes, if the first magnitude flag of the current scanned transform coefficient is positive, Then use context adaptive entropy decoding to decode the second magnitude flag of one of the currently scanned transform coefficients. For example, the device of claim 6, which is configured to calculate the absolute value of the quantization index based on a sum for a predetermined transformation coefficient, and the addend of the sum is formed by the following: the effective value flag of the predetermined transformation coefficient , The parity flag of the predetermined transform coefficient, the remainder of the predetermined transform coefficient, the first magnitude flag, and the second magnitude flag are twice the sum of one. For example, the device of claim 6, which is configured to calculate the absolute value of the quantization index based on a sum for a predetermined transformation coefficient, and the addend of the sum is formed by the following: the effective value flag of the predetermined transformation coefficient , The parity flag of the predetermined transform coefficient, the first degree flag of the predetermined transform coefficient, and the remainder of the predetermined transform coefficient twice the sum of the second degree flag. For example, the device of claim 6, which is configured to perform the decoding of the first magnitude flag and the second magnitude flag in a single second pass, and/or the one or more second The pass does not include the one or more first passes. For example, the device of claim 1, which is configured to execute each of the one or more second passes after each of the one or more first passes, and in the one or more first passes Each of the second passes is followed by each of the one or more third passes. Such as the device of claim 1, wherein the one or more first passes are included in the one or more second passes. Such as the device of claim 11, wherein the effective value flag, the parity flag, and the one or more magnitude flags of the transform coefficients are decoded in one pass. Such as the device of claim 1, which is configured to use context-adaptive entropy decoding for a current scanned transform coefficient in one of the one or more first passes The effective value flag is decoded, and after the first pass, in the one or more second passes, context adaptive entropy decoding is used to flag more than one magnitude of the transform coefficients A sequence is decoded, after the one or more second passes, after the one or more first passes In the other of them, equal-probability entropy decoding is used to decode the parity flags of the transform coefficients, and the quantization index of the transform coefficients is not zero. Such as the device of claim 1, wherein the transform coefficients of the block are divided into sub-blocks, and the transform coefficients are decoded sub-block by sub-block, so that the transform coefficients of one sub-block are in the next sub-block The transform coefficients have been decoded before. Such as the device of claim 1, wherein binary arithmetic decoding is used for entropy decoding, and the remainder is decoded by decoding one of the binary bits of an absolute value of the quantization index of the transform coefficients. For example, the device of claim 1, which is configured to dequantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depends on the quantization of the transform coefficients that precede the individual transform coefficients along a scanning order For index parity, select a reconstruction level set from multiple reconstruction level sets for the individual transform coefficient, and dequantize the individual transform coefficient to a level of the reconstruction level set, and the level is determined by the individual transform coefficient. The quantization index of the transform coefficient is indexed. For example, the device of claim 16, which is configured to use a state transition to perform the operation of selecting the reconstruction level set from a plurality of reconstruction level sets for the respective transform coefficient by the following operation: uniquely based on the The state transition is false for the individual transform coefficient Determine a state to select the reconstruction level set from the plurality of reconstruction level sets, depending on the parity of the quantization index of the respective transform coefficient, and update the state transition for the subsequent transform coefficient in the scan order The state. For example, the device of claim 17, which is configured to execute the one or more first passes, and/or the one or more second passes, and/or the one or more first passes along the scanning order Three times. For example, the device of claim 17, which is configured to perform state transitions between four different states. For example, the device of claim 16, which is configured to parameterize the multiple reconstruction level sets by means of a predetermined quantization step size and derive information about the predetermined quantization step size from the data stream. Such as the device of claim 17, wherein each of the multiple reconstruction level sets is composed of a multiple of a predetermined quantization step size, and the predetermined quantization step size is constant for the multiple reconstruction level sets. Such as the device of claim 16, wherein the number of reconstruction level sets in the multiple reconstruction level sets is two, and the multiple reconstruction level sets include a first reconstruction level set, which includes zero and one An even multiple of the predetermined quantization step size, and a second reconstruction level set, which includes zero and the predetermined quantization step size Odd multiples. For example, the device of request 22 is configured to select the first reconstruction level set for state values 0 and 1, and select the second reconstruction level set for state values 2 and 3. For example, the device of claim 1, which is configured to decode the effective value flag of a predetermined transform coefficient by using context-adaptive entropy decoding by selecting a context depending on the following: the predetermined inside the transform block The position of one of the transform coefficients. For example, the device of claim 1, which is configured to use context-adaptive entropy decoding to decode the effective value flag of a predetermined transform coefficient by the following operation: based on a local template surrounding the predetermined transform coefficient A set of flags decoded before the effective value flag of the predetermined transform coefficient of a set of adjacent transform coefficients determines a local activity, and a context is selected depending on the local activity. For example, the device of claim 25, which is configured to be one of the effective value flag, the parity flag, and the one or more size flags in order in a first pass. The flag is decoded so that the flag set includes the effective value flag, the parity flag, and the first size flag of the one or more size flags decoded for the set of adjacent transform coefficients , And is configured to calculate the activity based on a sum of the addends of each of the adjacent transform coefficients, the addends indicating the minimum hypothetical index for one of the adjacent transform coefficients Or a minimally assumed reconstruction level, the index or reconstruction level is based on It is determined on the first magnitude flag among the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients. For example, the device of claim 1, which is configured to use a state transition to perform the operation of selecting the reconstruction level set from a plurality of reconstruction level sets for the respective transform coefficient by the following operation: uniquely based on the The state transition assumes a state for the respective transform coefficient to select the reconstruction level set from the plurality of reconstruction level sets, depending on the parity of the quantization index of the respective transform coefficient, for the latter in the scan order Update the state of the state transition with a transform coefficient of, perform the one or more first passes along the scanning order, and use context-adaptive entropy decoding on a predetermined transform coefficient by selecting a context depending on the following The effective value flag is decoded: the state transition is based on the state assumed for the predetermined transform coefficient. For example, the device of claim 27, which is configured to decode the effective value flag of a predetermined transform coefficient by using context adaptive entropy decoding by selecting a context set containing the context depending on the state, and depending on the state The context is selected from the context set at a local activity surrounding one of the predetermined transformation coefficients or a coefficient position of the predetermined transformation coefficient. Such as the equipment of claim 28, which is equipped with targeted status States 0 and 1 select a first context set, for state 2, a second context set, and for state 3, a third context set. For example, the device of claim 1, which is configured to dequantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depends on the transform coefficients that precede the respective transform coefficient along a scanning order For the parity of the quantization index, a reconstruction level set is selected from a plurality of reconstruction level sets for the respective transform coefficient, and the respective transform coefficient is dequantized to a level of the reconstruction level set, and the level is determined by each The quantization index of the transform coefficient is indexed, the one or more first passes are performed along the scanning order, and the context-adaptive entropy decoding is used to determine the conversion of a predetermined transform coefficient by selecting a context depending on the following The effective value flag is decoded: the reconstruction level set selected for the predetermined transform coefficient. For example, the device of claim 1, which is configured to decode the parity flag of a predetermined transform coefficient by using context-adaptive entropy decoding by selecting a context depending on the following: a coefficient position of the predetermined transform coefficient . For example, the device of claim 1, which is configured to use context-adaptive entropy decoding to decode the parity flag of a predetermined transform coefficient by the following operation: based on a local template surrounding the predetermined transform coefficient The parity flag of the predetermined transform coefficient of a group of adjacent transform coefficients A previously decoded flag set determines a local activity, and/or determines a number of transform coefficients in the local template surrounding the predetermined transform coefficients, the reconstruction level of the predetermined transform coefficients is not zero, and depends Choose a context in the local activity and/or the number of transform coefficients. Such as the device of claim 32, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients. For example, the device of claim 32, which is configured to be one of the effective value flag, the parity flag, and the one or more size flags in a first pass. The flag is decoded so that the flag set includes the effective value flag, the parity flag, and the first size flag of the one or more size flags decoded for the set of adjacent transform coefficients , And is configured to calculate the activity based on a sum of the addends of each of the adjacent transform coefficients, the addends indicating the minimum hypothetical index for one of the adjacent transform coefficients Or a minimally assumed reconstruction level based on the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients The first degree flag is determined. Such as the device of claim 1, which is configured to use context-adaptive entropy decoding for one or more of the one or more magnitude flags of a predetermined transform coefficient by selecting a context depending on the following A first magnitude flag is decoded: a coefficient position of the predetermined transform coefficient. For example, the device of claim 1, which is configured to decode the first magnitude flag of a predetermined transform coefficient using context-adaptive entropy decoding by the following operation: A set of flags decoded before the first magnitude flag of the predetermined transformation coefficient of a set of adjacent transformation coefficients in a template determines a local activity, and/or determines a local activity in the local template surrounding the predetermined transformation coefficient A number of transform coefficients, the reconstruction level of the predetermined transform coefficients is not zero, and a context is selected depending on the local activity and/or the number of transform coefficients. Such as the device of claim 36, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients. For example, the device of claim 36, which is configured to be one of the effective value flag, the parity flag, and the one or more size flags in a first pass. The flag is decoded so that the flag set includes the effective value flag, the parity flag, and the first size flag of the one or more size flags decoded for the set of adjacent transform coefficients , And is configured to calculate the local activity based on a sum of the addends of each of the adjacent transformation coefficients, and the addend indications are used for one of the adjacent transformation coefficients, which is minimally assumed Index or a minimally assumed reconstruction level based on the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients It is determined by the first level flag. For example, the device of claim 1, which is configured to use a Columbus-Rice code and use a Rice parameter to decode the remainder of a predetermined transform coefficient. The Rice parameter depends on a part of the predetermined transform coefficient. A sum of the addends of each of a group of adjacent transform coefficients in the template, and the addend depends on a quantization index or a reconstruction level of each adjacent transform coefficient. A device for encoding a block of transform coefficients. The device is configured to use context-adaptive entropy coding to encode the current transform coefficient in one or more first passes of the transform coefficients. A valid value flag is coded, the valid value flag indicates whether a quantization index of the current transform coefficients is zero, and a parity flag of one of the transform coefficients is encoded, and the parity flag indicates the value of the transform coefficients. Together, in one or more second passes of scanning the transform coefficients including or not including one or more of the one or more first passes, the quantization is performed using context-adaptive entropy coding One of the transform coefficients whose index is not zero or multiple magnitude flags are coded, and in one or more third passes, the following are coded using equal-probability entropy coding, The remainder of the quantization index of the transform coefficients, the one or more magnitude flags of the transform coefficients are positive, and a sign of the quantization index of the transform coefficients, the quantization index of the transform coefficients Not zero. For example, the device of claim 40, which is configured to decode the effective value flag of the current scanned transform coefficient by using context adaptive entropy coding for a current scanned transform coefficient in a first pass And if the effective value flag indicates that the quantization index of the current scanned transform coefficient is not zero, then for the current scanned transform coefficient, use context adaptive entropy coding for the parity flag of the current scanned transform coefficient Mark to be coded. For example, the device of claim 40, which is configured to use context-adaptive entropy coding for a current scanned transform coefficient in one of the one or more first passes The parity flag is encoded, and in the other of the one or more first passes, for a current scanned transform coefficient, if the parity flag indicates the quantization index of the current scanned transform coefficient If it is an even number, for the current scanned transform coefficient, use context adaptive entropy coding to encode the effective value flag of the current scanned transform coefficient. Such as the equipment of claim 33, which is assembled with In the one or more second passes, for a current scanned transform coefficient, if the quantization index of the current scanned transform coefficient is not zero, context adaptive entropy coding is used for the current scanned transform coefficient One of the first size level flags is encoded. For example, the device of claim 43 is configured so that for a predetermined transform coefficient, the quantization index can be obtained according to a sum, and the addend of the sum is formed by: the effective value flag of the predetermined transform coefficient, The parity flag of the predetermined transform coefficient, the remainder of the predetermined transform coefficient and the sum of one of the first magnitude flags are twice the sum. For example, the device of claim 44, which is configured for a current scanned transform coefficient in the one or more second passes, if the first magnitude flag of the current scanned transform coefficient is positive, Then use context adaptive entropy coding to encode the second magnitude flag of one of the currently scanned transform coefficients. For example, the device of claim 45, which is configured to obtain the quantization index for a predetermined transform coefficient according to a sum, and the addend of the sum is formed by the following: the effective value flag of the predetermined transform coefficient, the The parity flag of the predetermined transform coefficient, the remainder of the predetermined transform coefficient, the first magnitude flag Twice the sum of one of the second size degree flags. For example, the device of claim 45, which is configured to obtain the quantization index for a predetermined transform coefficient according to a sum, and the addend of the sum is formed by the following: the effective value flag of the predetermined transform coefficient, the The parity flag of the predetermined transform coefficient, the first degree flag of the predetermined transform coefficient, and the remainder of the predetermined transform coefficient twice the sum of the second degree flag. For example, the device of claim 45 is configured to execute the coding of the first size flag and the second size flag in a single second pass, and/or one of them The one or more second passes do not include the one or more first passes. For example, the device of claim 33 is configured to execute each of the one or more second passes after each of the one or more first passes, and in the one or more first passes, Each of the second passes is followed by each of the one or more third passes. Such as the device of claim 40, wherein the one or more first passes are included in the one or more second passes. Such as the device of claim 50, wherein the effective value flag, the parity flag, and the one or more magnitude flags of the transform coefficients are decoded in one pass. For example, the device of claim 40, which is configured to use context-adaptive entropy coding for a current scanned transform coefficient in one of the one or more first passes The effective value flag is encoded, and after the first pass, in the one or more second passes, context adaptive entropy decoding is used to flag more than one magnitude of the transform coefficients A sequence is coded, and after the one or more second passes, in the other of the one or more first passes, the parity flag of the transform coefficients is marked with equal-probability entropy coding When encoding, the quantization index of the transform coefficients is not zero. Such as the device of claim 40, wherein the transform coefficients of the block are divided into sub-blocks, and the transform coefficients are coded sub-block by sub-block, so that the transform coefficients of one sub-block are in the next sub-block The transform coefficients have been coded before. Such as the device of claim 40, wherein binary arithmetic coding is used for entropy coding, and the remainder is coded by coding a binary bit of an absolute value of the quantization index of the transform coefficients. For example, the device of claim 40, which is configured to quantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depends on the quantization of the transform coefficient that precedes the individual transform coefficient along a scanning order The parity of the index, for the individual transform coefficients A set of reconstruction levels is selected from a set of reconstruction levels, and the respective transform coefficient is quantized to a level of the set of reconstruction levels, and the level is indexed by the quantization index of the respective transform coefficient. For example, the device of claim 55 is configured to use a state transition to perform the operation of selecting the reconstruction level set from a plurality of reconstruction level sets for the respective transform coefficient by the following operation: uniquely based on the The state transition assumes a state for the respective transform coefficient to select the reconstruction level set from the plurality of reconstruction level sets, depending on the parity of the quantization index of the respective transform coefficient, for the latter in the scan order A transform coefficient of updates the state of the state transition. For example, the device of claim 56, which is configured to execute the one or more first passes, and/or the one or more second passes, and/or the one or more second passes along the scanning order Three times. For example, the device of claim 56 is configured to perform state transitions between four different states. For example, the equipment of claim 55 is configured to parameterize the multiple reconstruction level sets by means of a predetermined quantization step size and derive information about the predetermined quantization step size from the data stream. Such as the device of claim 56, wherein each of the multiple reconstruction level sets is composed of a multiple of a predetermined quantization step size, and the predetermined quantization step size is constant for the multiple reconstruction level sets. Such as the device of claim 55, wherein the number of reconstruction level sets in the multiple reconstruction level sets is two, and the multiple reconstruction level sets include a first reconstruction level set, which includes zero and one An even multiple of a predetermined quantization step, and a second reconstruction level set, which includes zero and an odd multiple of the predetermined quantization step. For example, the device of request 61 is configured to select the first reconstruction level set for state values 0 and 1, and select the second reconstruction level set for state values 2 and 3. For example, the device of claim 40, which is configured to use context-adaptive entropy coding to encode the effective value flag of a predetermined transformation coefficient by selecting a context depending on the following: the predetermined within the transformation block The position of one of the transform coefficients. For example, the device of claim 40, which is configured to use context adaptive entropy coding to encode the effective value flag of a predetermined transform coefficient by the following operation: based on a local template surrounding the predetermined transform coefficient A set of flags decoded before the effective value flag of the predetermined transform coefficient of a set of adjacent transform coefficients determines a local activity, and a context is selected depending on the local activity. For example, the device of claim 64 is configured to sequentially in a first pass, one of the effective value flag, the parity flag, and the one or more size flags is the first size level The flags are encoded so that the flag set includes the effective value flag, the parity flag, and the first size flag of the one or more size flags decoded for the set of adjacent transform coefficients , And is configured to calculate the activity based on a sum of the addends of each of the adjacent transform coefficients, the addends indicating the minimum hypothetical index for one of the adjacent transform coefficients Or a minimally assumed reconstruction level based on the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients The first degree flag is determined. For example, the device of claim 40 is configured to use a state transition to perform the operation of selecting the reconstruction level set from a plurality of reconstruction level sets for the respective transform coefficient by the following operation: uniquely based on the The state transition assumes a state for the respective transform coefficient to select the reconstruction level set from the plurality of reconstruction level sets, depending on the parity of the quantization index of the respective transform coefficient, for the latter in the scan order Update the state of the state transition with a transform coefficient of, perform the one or more first passes along the scan order, and use context adaptivity by selecting a context depending on the following Entropy coding encodes the effective value flag of a predetermined transform coefficient: the state transition is for the state assumed by the predetermined transform coefficient. For example, the device of claim 66 is configured to encode the effective value flag of a predetermined transform coefficient by using context adaptive entropy coding by selecting a context set containing the context depending on the state, and depending on the state The context is selected from the context set at a local activity surrounding one of the predetermined transformation coefficients or a coefficient position of the predetermined transformation coefficient. For example, the device of request 67 is configured to select a first context set for states 0 and 1, a second context set for state 2, and a third context set for state 3. For example, the device of claim 40, which is configured to quantize the quantization index of each transform coefficient whose quantization index is not zero by the following operation: depends on the quantization of the transform coefficient that precedes the individual transform coefficient along a scanning order For index parity, select a reconstruction level set from multiple reconstruction level sets for the individual transform coefficient, and dequantize the individual transform coefficient to a level of the reconstruction level set, and the level is determined by the individual transform coefficient. The quantization index of the transform coefficient is indexed, the one or more first passes are executed along the scan order, and the context-adaptive entropy coding is used to select a context depending on the following to apply the context adaptive entropy coding to the predetermined transform coefficient The effective value flag is coded: the reconstruction level set selected for the predetermined transform coefficient. Such as the device of claim 40, which is configured to use context-adaptive entropy coding to encode the parity flag of a predetermined transform coefficient by selecting a context depending on the following: a coefficient position of the predetermined transform coefficient . For example, the device of claim 40, which is configured to use context adaptive entropy coding to encode the parity flag of a predetermined transform coefficient by the following operation: based on a local template surrounding the predetermined transform coefficient A set of flags decoded before the parity flag of the predetermined transform coefficient of a set of adjacent transform coefficients determines a local activity, and/or determines a number of transform coefficients in the local template surrounding the predetermined transform coefficient , The reconstruction level of the predetermined transform coefficients is not zero, and a context is selected depending on the local activity and/or the number of transform coefficients. Such as the device of claim 71, which is configured to select the context depending on a difference between the local activity and the number of transform coefficients. For example, the device of claim 71, which is configured to be one of the effective value flag, the parity flag, and the one or more size flags in a first pass. The flags are encoded so that the flag set includes the effective value flag, the parity flag, and the first size flag of the one or more size flags decoded for the set of adjacent transform coefficients , And are combined to be based on these adjacent changes The activity is calculated by converting a sum of one of the addends of each of the coefficients, the addends indicating a minimally assumed index or a minimally assumed reconstruction level for one of the adjacent transformation coefficients, The index or reconstruction level is determined based on the first magnitude flag among the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients. Such as the device of claim 40, which is configured to use context-adaptive entropy coding to first size one of the one or more size flags of a predetermined transform coefficient by selecting a context depending on the following The degree flag is coded: the position of one of the predetermined transform coefficients. For example, the device of claim 40 is configured to use context adaptive entropy coding to encode the first magnitude flag of a predetermined transform coefficient by the following operation: A set of flags decoded before the first magnitude flag of the predetermined transformation coefficient of a group of adjacent transformation coefficients in a template determines a local activity, and/or determines a local activity around the predetermined transformation coefficient in the local template A number of transform coefficients, the reconstruction level of the predetermined transform coefficients is not zero, and a context is selected depending on the local activity and/or the number of transform coefficients. For example, the device of claim 75 is configured to select the upper and lower values depending on the difference between the local activity and the number of transform coefficients Arts. For example, the device of claim 75 is configured to sequentially in a first pass, one of the effective value flag, the parity flag, and the one or more size flags is the first size level The flags are encoded so that the flag set includes the effective value flag, the parity flag, and the first size flag of the one or more size flags decoded for the set of adjacent transform coefficients , And is configured to calculate the local activity based on a sum of the addends of each of the adjacent transformation coefficients, and the addend indications are used for one of the adjacent transformation coefficients, which is minimally assumed Index or a minimally assumed reconstruction level based on the effective value flag, the parity flag, and the one or more magnitude flags decoded for the adjacent transform coefficients It is determined by the first level flag. For example, the device of claim 40, which is configured to use a Columbus-Rice code and use a Rice parameter to encode the remainder of a predetermined transform coefficient, the Rice parameter depends on a part around the predetermined transform coefficient A sum of the addends of each of a group of adjacent transform coefficients in the template, and the addend depends on a quantization index or a reconstruction level of each adjacent transform coefficient. A method for decoding a block of transform coefficients. The method is configured to use context-adaptive entropy decoding to decode the transform coefficients in one or more first passes of the transform coefficients. One of the effective value flags is decoded, and the effective value flag indicates the transformations Whether the quantization index of one of the coefficients is zero, decode a parity flag of one of the transform coefficients, and the parity flag indicates that one of the transform coefficients is parity, in the first pass including or not including the one or more In one or more second passes of scanning one or more of the transform coefficients, use context-adaptive entropy decoding to decode one or more magnitude flags of the transform coefficients whose quantization index is not zero In one or more third passes, use equal-probability entropy decoding to decode each of the following, the remainder of the quantization index of the transform coefficients, and the one or more magnitude flags of the transform coefficients Is positive, and a sign of the quantization index of the transform coefficients, the quantization index of the transform coefficients is not zero. A method for encoding a block of transform coefficients. The method is configured to use context-adaptive entropy coding to encode the current transform coefficient in one or more first passes of scanning the transform coefficients. A valid value flag is coded, the valid value flag indicates whether a quantization index of the current transform coefficients is zero, and a parity flag of one of the transform coefficients is encoded, and the parity flag indicates the value of the transform coefficients. Together, in one or more second passes of scanning the transform coefficients including or not including one or more of the one or more first passes, the quantization is performed using context-adaptive entropy coding Index is not One or more magnitude flags of zero transform coefficients are coded. In one or more third passes, the following are coded using equal-probability entropy coding. The remainder of the quantization index of the transform coefficients , The one or more magnitude flags of the transform coefficients are positive, and the quantization index of the transform coefficients is a sign, and the quantization index of the transform coefficients is not zero. A computer program used to send instructions to a computer, and the computer executes the computer program to perform a method such as request item 79 or 80. A data stream generated by a method such as request item 80.

Images